Agent selection service for the chemical industry

Abstract

A method for providing target chemical product associated with chemical product data, the chemical product data characterizing the target chemical product and / or one or more properties of the chemical product, the method comprising: providing a request for providing the target chemical product, wherein the request includes unstructured data and is associated with an indication regarding the target chemical product; providing functional specification data associated with one or more functions of one or more operating engines for providing the chemical product data; providing a selection task, including unstructured data, associated with the request and the functional specification data to a selection model to generate operational input data, wherein the selection model is configured to select at least one operating engine based on the request and the functional specification data, wherein the operational input data indicates the at least one selected operating engine, and wherein the operational input data is suitable for triggering the selected operating engine; providing the operational input data to the at least one selected operating engine to generate the chemical product data, wherein the selected operating engine is configured to provide at least a portion of the chemical product data in response to providing the operational input data; and providing the generated chemical product data to provide a target chemical product having the one or more target properties.

Description

Technical Field

[0001] This disclosure relates to chemical production, particularly customized chemical production to meet customer needs, and includes a method for generating chemical product data characterizing a chemical product having one or more target properties, a method for producing and / or processing a target chemical product associated with chemical product data (the chemical product data characterizing the target chemical product and / or one or more properties of the chemical product), an apparatus, the use of one or more data-driven models, the use of task instructions, the use of chemical product data, and a method for producing a chemical product having one or more target properties. Background Technology

[0002] Chemical production networks are complex production systems used to produce hundreds of different chemical products with varying properties. Therefore, providing chemical products with customized properties is challenging. Summary of the Invention

[0003] In one aspect, this disclosure relates to a method, particularly a computer-implemented method, for generating chemical product data characterizing chemical products having one or more target properties, comprising:

[0004] The request to provide a chemical product having one or more target properties is received, particularly preferably via an interface such as a user interface, wherein the request includes unstructured data and is associated with one or more target properties of the chemical product.

[0005] Obtaining, and particularly preferably receiving via an interface such as a user interface, functional specification data related to one or more functions of one or more operating engines for providing chemical products having the one or more target characteristics.

[0006] Specifically, the processing device provides one or more input data structures related to the input data suitable for being provided to the one or more operating engines.

[0007] Specifically, the processing device provides task instructions, including unstructured data, related to the request, the one or more input data structures, and the functional specification to one or more data-driven models, wherein the one or more data-driven models are configured to generate operational input data for one or more selected operation engines in response to one or more provided target characteristics, wherein the operational input data includes structured data for triggering the selected operation engine.

[0008] Specifically, the processing device provides the operational input data to the at least one selected operational engine to generate the chemical product data, wherein the operational engine is configured to generate at least a portion of the chemical product data in response to the provision of the operational input data.

[0009] Specifically, the processing equipment and / or interface provide data on the generated chemical products to produce chemical products having one or more of the target properties.

[0010] On the other hand, this disclosure relates to a method for producing and / or processing a target chemical product associated with chemical product data characterizing the target chemical product and / or one or more properties of the chemical product, the method comprising:

[0011] The request to provide a chemical product having one or more target properties is received, particularly preferably via an interface such as a user interface, wherein the request includes unstructured data and is associated with one or more target properties of the chemical product.

[0012] Obtaining, and particularly preferably receiving via an interface such as a user interface, functional specification data related to one or more functions of one or more operating engines for providing chemical products having the one or more target characteristics.

[0013] Specifically, the processing device provides one or more input data structures related to the input data suitable for being provided to the one or more operating engines.

[0014] Specifically, the processing device provides task instructions, including unstructured data, related to the request, the one or more input data structures, and the functional specification to one or more data-driven models, wherein the one or more data-driven models are configured to generate operational input data for one or more selected operation engines in response to one or more provided target characteristics, wherein the operational input data includes structured data for triggering the selected operation engine.

[0015] Specifically, the processing device provides the operational input data to the at least one selected operational engine to generate the chemical product data, wherein the operational engine is configured to generate at least a portion of the chemical product data in response to the provision of the operational input data.

[0016] Specifically, the processing equipment and / or interface provide data on the generated chemical products to produce chemical products having one or more of the target properties.

[0017] Optionally, a target chemical product having one or more of the target properties may be produced and / or processed.

[0018] On the other hand, this disclosure relates to the use of mission instructions as described herein for processing requests for providing chemical product data to produce and / or process target chemical products as described herein.

[0019] On the other hand, this disclosure relates to the use of one or more data-driven models, as described herein, to provide chemical product data for the production and / or processing of target chemical products.

[0020] On the other hand, this disclosure relates to an apparatus comprising: a processor; and a memory storing instructions that, when executed by the processor, configure the apparatus to perform any of the methods presented herein.

[0021] On the other hand, this disclosure relates to the use of chemical product data obtained by any of the methods described herein for the production and / or processing of chemical products.

[0022] On the other hand, this disclosure relates to the use of task instructions according to any of the methods described herein for processing a request to provide a chemical product having one or more target properties according to any of the methods described herein.

[0023] On the other hand, this disclosure relates to a method for producing a chemical product having one or more target properties, the method comprising:

[0024] A request is provided for the production of a chemical product having one or more target properties, wherein the request includes unstructured data and is associated with one or more target properties of the chemical product.

[0025] Provide functional specification data related to one or more functions of one or more operating engines used to provide chemical products having one or more target properties.

[0026] Provide one or more input data structures that are suitable for the input data provided to the one or more operating engines.

[0027] Task instructions, including unstructured data, related to the request, the one or more input data structures, and the functional specification are provided to one or more data-driven models, wherein the one or more data-driven models are configured to generate operational input data for one or more selected operation engines in response to one or more provided target characteristics, wherein the operational input data includes structured data for triggering the selected operation engine.

[0028] The operational input data is provided to the at least one selected operational engine to generate chemical product data characterizing a chemical product having the one or more target properties, wherein the operational engine is configured to generate at least a portion of the chemical product data in response to the provision of the operational input data.

[0029] The chemical product is produced based on the data of this chemical product. Example

[0030] Any disclosures, embodiments, and examples described herein relate to the methods, systems, apparatuses, chemical products, and computer elements listed above and below. Advantageously, the benefits provided by any embodiments and examples also apply to all other embodiments and examples.

[0031] In the following sections, the terminology and / or technical fields of this disclosure as used herein will be outlined by way of definition and / or examples. Where examples are given, it should be understood that this disclosure is not limited to the examples described.

[0032] These and other objectives are addressed by the subject matter of the independent claims, and will become apparent upon reading the following description. The dependent claims relate to embodiments of the content of this disclosure.

[0033] Chemical products are the starting materials for a variety of different end products. Therefore, chemical products must offer a range of varying properties tailored to the desired end product. The production of chemical products begins with raw materials, which are processed through one or more processing steps, including, for example, chemical reactions in a reactor and purification steps. Typically, chemical products are obtained from two or more chemical reactions that alter the chemical structure of the reactants and thus the properties of the chemical product. For example, liquids (such as monoethylene glycol) and solids (such as terephthalic acid) can be converted to produce polyesters. Polyesters are functional polymers with different properties depending on the reactants and reaction conditions. This can result in polyesters with different properties. Therefore, the chemical reactions used to produce polyesters need to be tailored to the desired properties of the polyester. Polyesters can be used in numerous fields, such as clothing or packaging. Therefore, the challenge lies in providing thousands of different chemical products to hundreds of customers, obtained from chemical production networks with multiple production steps to achieve different and customized properties. The properties of chemical products are highly dependent on their chemical structure. Even a small change in the orientation of a subgroup in a molecule with hundreds of atoms can lead to different chemical properties. Therefore, the relationship between the chemical structure and properties of chemical products is complex and difficult to control.

[0034] Processing requests for chemical products possessing one or more target properties allows for the acquisition of the target chemical product in response to the receipt of an unstructured request. This is particularly advantageous because indications regarding the target chemical product can have various different formats. For example, ethylene could be an IUPAC name, while the same chemical compound could be associated with the common name ethylene. Furthermore, companies sell chemical products under their established product names. Providing task instructions to one or more data-driven models to provide a digital representation of the chemical structure of the target chemical product as indicated by the provided indications, and providing chemical product data based on this digital representation, enables the efficient and robust determination of chemical product data for the production of the target chemical product upon receipt of an unstructured request (e.g., including the common name or other synonyms of the chemical product). This leads to the reliable processing and / or production of the chemical product associated with the target product.

[0035] In embodiments, chemical product data may be associated with a digital representation of the chemical structure of a chemical product having one or more target properties. The chemical product data may be associated with and / or obtainable from one or more target properties. The chemical product data may depend on one or more target properties provided. Specifically, the chemical product data may include a digital representation of the chemical structure of a chemical product having one or more target properties. Further, the chemical product data may include unstructured data, such as string data. The chemical product data may include machine instructions for providing a chemical product having one or more target properties. The machine instructions may be provided for producing and / or processing a chemical product having one or more target properties.

[0036] In embodiments, the indication of a target chemical product may be data indicating the chemical product. The indication of a target chemical product may be related to the chemical product and / or the characteristics of the chemical product. The indication of a target chemical product may identify the chemical product. The indication of a target chemical product may be related to the chemical structure of the chemical product. Therefore, the indication of a target chemical product may be related to a designation associated with the chemical product.

[0037] In embodiments, the numerical representation of a chemical structure can refer to a machine-readable and / or machine-interpretable representation of the chemical structure of a chemical product. The numerical representation of a chemical structure can indicate the chemical structure of a chemical product. A chemical structure can specify one or more atoms associated with a chemical product. Specifying one or more atoms can mean specifying one or more elements associated with one or more atoms. Further, the numerical representation of a chemical structure can indicate a relationship between one or more atoms, preferably an interaction between one or more atoms, and most preferably one or more bonds between one or more atoms. Alternatively or additionally, the numerical representation of a chemical structure can indicate the spatial arrangement of one or more atoms, preferably relative to a predefined point and / or relative to at least one atom among one or more atoms. The numerical representation of a chemical structure can include string data, particularly indicating one or more elements associated with one or more atoms, and / or numerical data, particularly indicating a relationship between one or more atoms.

[0038] In embodiments, functional specification data may be associated with one or more functions of one or more operating engines. Functional specification data may include functional specifications associated with one or more operating engines. Functional specification data may instruct one or more operating engines on the processing of operational input data, particularly the processing of operational output data. Functional specifications associated with one or more operating engines may indicate one or more functions associated with one or more operating engines, particularly one or more functions performed by one or more operating engines. One or more functions may include at least one function associated with providing a chemical product having one or more target properties. This at least one function may configure one or more operating engines (particularly selected operating engines) to provide a chemical product having one or more target properties. Further, functional specifications may indicate input data and / or output data associated with one or more operating engines. Output data associated with an operating engine may be operational output data. Chemical product data may include at least a portion of the operational output data. Functional specification data may include unstructured data.

[0039] In embodiments, one or more input data structures may indicate one or more input data formats suitable for being provided to one or more operating engines. One or more input data structures may refer to an arrangement of input data (particularly input data of one or more operating engines). Operational input data may be input data of one or more operating engines. The input data structure may indicate an arrangement of one or more target characteristics (optionally a target type of chemical product and / or a target application area associated with the chemical product). Operational input data associated with one or more input data structures may be provided to one or more operating engines (particularly at least one selected operating engine).

[0040] In embodiments, operational input data may be provided to one or more operational engines (particularly selected operational engines). Operational input data may include structured data. Operational input data may be associated with, and particularly include, an indication regarding a target chemical product. Operational input data may be obtained from and / or may depend on the indication regarding the target chemical product. Operational input data may be associated with a data structure associated with the selected operational engine. One or more operational engines may be configured to receive operational input data and generate at least a portion of chemical product data (particularly operational output data) based on the operational input data. Operational input data may be associated with a digital representation of the chemical structure of the chemical product, and particularly may include a digital representation of the chemical structure of the chemical product. Operational input data may be obtained from and / or may depend on the one or more target characteristics. Operational output data may be associated with at least a portion of the chemical product data, and particularly may relate to, and more preferably include, at least a portion of the chemical product data.

[0041] In an embodiment, a request to provide a chemical product having one or more target properties may include unstructured data. The request may be associated with one or more target properties of the chemical product. The request may include one or more target properties. Further, the request may include user instructions for providing a chemical product having one or more target properties. The request and / or user instructions may include string data. The user instructions may indicate a function to be performed by a selected operating engine. The request and / or target properties may include numerical data. The request may further include instructions regarding the target chemical product. Instructions regarding the target chemical product may be adapted to select subgroups of the chemical product. Instructions regarding the chemical product may include the target type of the chemical product, the target application area of ​​the chemical product, and / or the quantity associated with the chemical product. The request may be provided and / or received via a user interface.

[0042] In embodiments, one or more target characteristics may be one or more characteristics required for processing chemical products (particularly the final product). One or more target characteristics may include one or more chemical characteristics, one or more physical characteristics, one or more environmental properties, and / or one or more biological characteristics. Target characteristics may include user-specific target characteristics. User-specific target characteristics may be provided and / or received via a user interface. Target characteristics may be provided by a chemical product processing facility. The chemical product processing facility may trigger the receipt of a request associated with one or more target characteristics.

[0043] In embodiments, environmental attributes may include at least one of the following: emission data for the chemical product, recyclable content of the chemical product, bio-based content of the chemical product, renewable bioavailability of the chemical product, chemical product declaration data, chemical product safety data, or a combination thereof. Emission data may include any data related to the environmental footprint. An environmental footprint may refer to an entity and its associated environmental footprint. An environmental footprint may be entity-specific. For example, an environmental footprint may relate to a chemical product, a company, a process such as a manufacturing process, raw materials or base substances, chemical products or materials, components, component assemblies, final products, combinations thereof, or additional entity-specific relationships. Emission data may include data related to the carbon footprint of the chemical product. Emission data may include data related to, for example, greenhouse gas emissions released during the production of the chemical product. Emission data may include data related to greenhouse gas emissions. Greenhouse gas emissions can include, for example, emissions of carbon dioxide (CO2), methane (CH4), nitrous oxide (N2O), hydrofluorocarbons (HFCs), perfluorocarbons (PFCs), sulfur hexafluoride (SF6), nitrogen trifluoride (NF3), and combinations thereof, as well as other emissions. Emissions data can include data related to greenhouse gas emissions generated by the entity or company's own operations (production, power plants, and waste incineration). Scope 2 includes emissions generated from the production of energy supplied externally. Scope 3 includes all other emissions generated along the value chain. Specifically, this includes greenhouse gas emissions from raw materials obtained from suppliers. Product carbon footprint (PCF) is the sum of greenhouse gas emissions and removals generated by consecutive and interrelated process steps associated with a specific product. Cradle-to-gate PCF aggregates greenhouse gas emissions based on selected process steps: from resource extraction to the product leaving the company's plant gates. This type of PCF is referred to as partial PCF. To achieve such aggregation, each company providing any product must be able to provide the contribution of each of its products to Scope 1 and Scope 2 of the PCF as accurately as possible, and obtain reliable and consistent PCF data for the energy purchased (Scope 2) and its raw materials (Scope 3).

[0044] In embodiments, chemical properties can be properties established by altering the chemical structure (particularly the chemical structure of a material). Chemical products can be obtained by altering the chemical structure of one or more reactants. The chemical properties of a chemical product can include properties associated with the chemical reaction of the chemical product. Examples include reactivity, electronegativity, etc. Physical properties can be one of the following: mechanical properties, electrical properties, optical properties, thermal properties, etc. For example, physical properties can include one or more of the following: density, scratch resistance, electrical conductivity, color, absorbency, heat capacity, etc. Biological properties can include properties related to the activities of living organisms.

[0045] In embodiments, task instructions may include at least a portion of a request, at least a portion of functional specification data, and / or at least a portion of one or more input data structures. Task instructions can be generated by combining at least a portion of the request, at least a portion of the functional specification data, and / or at least a portion of one or more input data structures. Task instructions may be provided to one or more data-driven models. The one or more data-driven models may be configured to select at least one operating engine from one or more operating engines. The selected operating engine may be associated with one or more functions for providing a chemical product having one or more target properties, particularly for providing a digital representation of the chemical structure of a chemical product having one or more target properties. In embodiments, task instructions may include selection task instructions and structural task instructions. Selection task instructions can be generated by combining at least a portion of the request and functional specification data. Selection task instructions may be provided to a selection model for selecting at least one operating engine. The selection model may provide model output data in response to being provided with a selection task instruction. The model output data may be associated with and / or indicate the selected operating engine and at least a portion of the request (particularly one or more target properties). Structural task instructions can be generated by merging at least a portion of the model output data and one or more input data structures. Structured task instructions can be provided to a structured model for generating operational input data (specifically, structured operational input data). The structured model can then provide operational input data in response to the provided structured task instructions.

[0046] In embodiments, providing task instructions may include mapping task instructions to numerical representations of task instructions. The numerical representation of task instructions may be a vectorized task instruction. Therefore, the terms vectorized task instruction and numerical representation of task instructions are used interchangeably. The numerical representation of task instructions may include tensors, such as matrices and / or vectors. One or more data-driven models may be configured to map the numerical representation of task instructions to numerical representations of operational input data, and vice versa. The numerical representation of task instructions may be associated with at least a portion of a request, at least a portion of functional specification data, and / or at least a portion of one or more input data structures, and / or may represent at least a portion of the request, at least a portion of the functional specification data, and / or at least a portion of the one or more input data structures. The numerical representation of task instructions may be obtained by passing the task instructions through one or more embedding layers. One or more data-driven models may include one or more embedding layers. One or more embedding layers may be configured to map unstructured data to structured numerical representations (particularly numerical representations of data). The numerical representation of task instructions may indicate and / or may depend on sequences and / or strings of data associated with one or more elements of the task instruction. The sequence can be encoded by positional encoding of the numerical representation of the task instructions. Therefore, the numerical representation of the task instructions can be mapped to the numerical representation of the task instructions and the relationship between two or more parts of the task instructions. The numerical representation of the task instructions and the relationship between two or more parts of the task instructions can be mapped to contextual task instructions. Positional encoding can be performed before the task instructions are provided to one or more data-driven models, or by processing one or more data-driven models (particularly one or more embedding layers). Alternatively, the encoder's self-attention mechanism can include relative positional encoding as described in 1803.02155.pdf (arxiv.org). The numerical representation of the task instructions can be associated with, and particularly include, structured numerical data (particularly tensors) related to requests, functional specification data, and one or more input data structures. Specifically, the numerical representation of the task instructions can represent the task instructions, preferably representing at least a portion of requests, one or more input data structures, and / or functional specification data. The numerical representation of the task instructions can be associated with a smaller amount of data than the task instructions themselves. Further, the numerical representation of the task instructions can be processed by one or more matrix operations associated with one or more data-driven models. Therefore, the numerical representation of task instructions can be processed more quickly through matrix operations of one or more data-driven models. The numerical representation of task instructions can be a structured numerical representation of the task instructions, including unstructured data (especially unstructured data associated with machine-processable numerical formats, preferably floating-point formats).The structured numerical representation of task instructions can be efficiently processed by one or more data-driven models, saving significant computational resources that would otherwise be used to process unstructured requests.

[0047] One or more data-driven models may include one or more encoder blocks and / or one or more decoder blocks for mapping numerical representations of task instructions to contextual task instructions. Contextual task instructions may be associated with numerical representations of task instructions. Contextual task instructions may be numerical representations of task instructions processed by one or more matrix operations associated with one or more data-driven models. One or more encoder blocks and / or one or more decoder blocks may be configured to map numerical representations of task instructions to contextual task instructions, particularly by considering sequences and / or string data of one or more elements associated with the task instructions. Contextual task instructions may be associated with structured numerical data (particularly tensors), and particularly include such structured numerical data. Contextual task instructions may represent sequences of elements associated with task instructions and relationships between elements in the sequence. Relationships between elements can be obtained by applying one or more matrix operations to the numerical representation of the task instructions. Elements may include at least a portion of words, numbers, symbols, etc. Thus, one or more data-driven models can understand relationships between elements (e.g., words in text). This improves the mapping between task instructions and operational input data. Therefore, the operational engine can be operated on more efficiently to process requests.

[0048] Furthermore, one or more data-driven models may include one or more encoder outputs, particularly where one or more data-driven models may include one or more encoder blocks. Further, one or more data-driven models may include one or more decoder outputs, particularly where one or more data-driven models may include one or more decoder blocks. Additionally or alternatively, one or more encoder outputs and / or one or more decoder outputs may be configured to map context task instructions to multiple confidence scores associated with multiple elements (particularly a distribution of confidence scores associated with multiple elements). At least a portion of the multiple elements may be associated with, and particularly included in, operational input data. One or more encoder blocks and / or decoder blocks may generate a distribution of confidence scores associated with the multiple elements (including one or more elements associated with the operational input data). The operational input data can be determined by selecting one or more elements associated with the operational input data based on one or more confidence scores associated with one or more elements. In embodiments, one or more encoder outputs and / or one or more decoder outputs may be configured to select one or more elements associated with the operational input data based on one or more confidence scores associated with one or more elements. Selecting one or more elements associated with operational input data based on one or more confidence scores associated with one or more elements may include receiving a range of confidence scores and selecting one or more elements associated with one or more confidence scores within that range. One or more encoder blocks and / or one or more encoder outputs and / or one or more decoder outputs may map the numerical representation of a task instruction to the operational input data, preferably to a distribution of confidence scores associated with multiple elements including one or more elements associated with the operational input data. One or more decoder blocks and one or more decoder outputs may map the numerical representation of a task instruction to the operational input data, preferably to a distribution of confidence scores associated with multiple elements including one or more elements associated with the operational input data.

[0049] In embodiments, unstructured data (particularly unstructured requests), unstructured task instructions, unstructured functional specification data, unstructured model output data, and / or unstructured chemical product data may include string data and / or sequences of one or more elements. Elements may include numbers, letters, symbols, etc. Humans need to understand and intervene in machine decisions, especially in domains with high security requirements where humans are domain experts. This disclosure allows for the conversion of machine-interpretable data into human-interpretable data. Furthermore, it enables intervention and / or triggering using human-interpretable data input. Therefore, this disclosure enables trustworthy AI.

[0050] In embodiments, generating operational input data for one or more selected operational engines for one or more provided target characteristics may include selecting at least one operational engine from one or more operational engines by one or more data-driven models based on task instructions (particularly request and functional specification data), and / or structuring the task instructions by one or more data-driven models to generate operational input data. At least one of the one or more data-driven models may be configured to select at least one operational engine from one or more operational engines based on task instructions (particularly request and functional specification data) and to structure the task instructions to generate operational input data. Alternatively, one or more data-driven models may include two or more data-driven models. The two or more data-driven models may be configured to perform at least one of the following: selecting at least one operational engine from one or more operational engines based on task instructions (particularly request and functional specification data), or structuring the task instructions for each data-driven model to generate operational input data. Task-specific models are more accurate than general models because they can be customized to satisfy one or more predefined functions. The two-preference processing of the request allows for reliable generation of operational input data. This, in turn, improves the robustness of generating chemical product data and thus improves the robustness of providing chemical products with one or more target characteristics.

[0051] In an embodiment, one or more data-driven models may be task-specific, partially task-independent, or task-independent. The selection model, validation model, structure model, and / or processing model may be data-driven models. In an embodiment, the selection model, validation model, structure model, and / or processing model may be task-specific, partially task-independent, or task-independent. One or more task-specific models may be configured to perform one of the following for each model:

[0052] Choose at least one operation engine from one or more operation engines, or

[0053] Structure task instructions to generate operational input data, or

[0054] Whether the data structures associated with the operation input data correspond to the input data structures associated with the selected operation engine, or

[0055] Unstructured data is generated in response to task instructions that are provided with at least some structured processing.

[0056] One or more partially independent data-driven models can be configured to perform one or more of the following for each model:

[0057] Choose at least one operation engine from one or more operation engines, or

[0058] Structure task instructions to generate operational input data, or

[0059] Whether the data structures associated with the operation input data correspond to the input data structures associated with the selected operation engine, or

[0060] Unstructured data is generated in response to task instructions that are provided with at least some structured processing.

[0061] In particular, at least one of one or more partially independent models can be configured to perform at least two of the following for each model:

[0062] Choose at least one operation engine from one or more operation engines, or

[0063] Structure task instructions to generate operational input data, or

[0064] Whether the data structures associated with the operation input data correspond to the input data structures associated with the selected operation engine, or

[0065] Unstructured data is generated in response to task instructions that are provided with at least some structured processing.

[0066] In an embodiment, one or more unrelated models may include at least one data-driven model configured to perform the following:

[0067] Choose at least one operation engine from one or more operation engines, and

[0068] Structure task instructions to generate operational input data, and

[0069] Classify whether the data structures associated with the operation input data correspond to the input data structures associated with the selected operation engine, and

[0070] Unstructured data is generated in response to task instructions that are provided with at least partial structured processing. In embodiments, one or more data-driven models may include structured data-driven models, selection models, validation models, and / or processing models. The advantage of at least partially independent models is that fewer models are required when generating chemical product data. Therefore, computational resources used to build and maintain multiple models can be saved or used to obtain more accurate at least partially independent models. In turn, using task-specific models allows for more accurate and robust generation of operational input data. This, in turn, improves the robustness of generating chemical product data and thus improves the robustness of providing chemical products with one or more target properties.

[0071] In an embodiment, providing task instructions to one or more data-driven models may include:

[0072] The selection task, including unstructured data, related to the request and the functional specification data is provided to the selection model, which is configured to generate model output data related to the selected operating engine and the one or more target features. The model output data includes unstructured data, and...

[0073] The generated model output data is provided to the structural model, which is configured to generate operational input data based on the model output data.

[0074] Specifically, providing task instructions to one or more data-driven models can include:

[0075] Selection task instructions related to request and functional specification data are provided to a selection model for selecting at least one operational engine from one or more operational engines. These selection task instructions may include unstructured data, and the selection model may be configured to generate model output data related to the selected operational engine and one or more target characteristics.

[0076] Structured task instructions, relating to model output data and one or more input data structures, are provided to a structured model for generating operational input data. This structured model can be configured to generate operational input data associated with a selected operational engine and one or more target characteristics, according to the structured task instructions. The structured task instructions may be associated with instructions configuring the structured model to generate structured operational input data. Selection task instructions may be associated with instructions configuring a selected model to generate structured model output data. Selection task instructions may include at least a portion of a request (particularly one or more target characteristics). Further, selection task instructions may be derived from a request, and / or the request may depend on one or more target characteristics. Selection task instructions may further include at least a portion of functional specification data. Model output data may be associated with at least one selected operational engine and / or one or more target characteristics. Preferably, the model output data may indicate at least one selected operational engine. Model output data may include one or more target characteristics. Dividing the task into two tasks allows for the use of task-specific data-driven models with one or more predefined functions. Typically, task-specific models are more accurate than general models because they can be customized to satisfy one or more predefined functions. The two requested biased processing allows for reliable generation of operational input data. This, in turn, improves the robustness of generating chemical product data, and thus improves the robustness of providing chemical products with one or more target properties.

[0077] In an embodiment, providing task instructions to one or more data-driven models may include:

[0078] The selection task, including unstructured data, related to the request and the functional specification data is provided to the selection model, which is configured to generate model output data related to the selected operating engine and the one or more target features. The model output data includes unstructured data, and...

[0079] The generated model output data is provided to the structural model, which is configured to generate operational input data based on the model output data.

[0080] Specifically, providing task instructions to one or more data-driven models can include:

[0081] Selection task instructions related to request and functional specification data are provided to a selection model for selecting at least one operational engine from one or more operational engines. These selection task instructions may include unstructured data, and the selection model may be configured to generate model output data related to the selected operational engine and one or more target characteristics.

[0082] Structured task instructions, relating to model output data and one or more input data structures, are provided to a structured model for generating operational input data. This structured model can be configured to generate operational input data associated with a selected operational engine and one or more target characteristics, according to the structured task instructions. The structured task instructions may be associated with instructions configuring the structured model to generate structured operational input data. Selection task instructions may be associated with instructions configuring a selected model to generate structured model output data. Selection task instructions may include at least a portion of a request (particularly one or more target characteristics). Further, selection task instructions may be derived from a request, and / or the request may depend on one or more target characteristics. Selection task instructions may further include at least a portion of functional specification data. Model output data may be associated with at least one selected operational engine and / or one or more target characteristics. Preferably, the model output data may indicate at least one selected operational engine. Model output data may include one or more target characteristics. Dividing the task into two tasks allows for the use of task-specific data-driven models with one or more predefined functions. Typically, task-specific models are more accurate than general models because they can be customized to satisfy one or more predefined functions. The two requested biased processing allows for reliable generation of operational input data. This, in turn, improves the robustness of generating chemical product data, and thus improves the robustness of providing chemical products with one or more target properties.

[0083] In an embodiment, providing a task selection instruction to the selection model may include mapping the task selection instruction to a numerical representation of the task selection instruction. The selection model may be configured to map the numerical representation of the task selection instruction to a numerical representation of the model output data, and vice versa.

[0084] The selection model may include one or more encoder blocks and / or one or more decoder blocks for mapping the numerical representation of the selection task instruction to the numerical representation of the model output data. The one or more encoder blocks and / or one or more decoder blocks may be configured to map the numerical representation of the selection task instruction to the numerical representation of the model output data, particularly by considering a sequence and / or string of data of one or more elements associated with the selection task instruction. Further, the selection model may include one or more encoder outputs, particularly where the selection model may include one or more encoder blocks. Further, the selection model may include one or more decoder outputs, particularly where the selection model may include one or more decoder blocks.

[0085] The selection model may include one or more encoder blocks and / or one or more decoder blocks for mapping a numerical representation of a selection task instruction to a contextual selection task instruction. The contextual selection task instruction may be associated with a numerical representation of the selection task instruction. The contextual selection task instruction may be a numerical representation of the selection task instruction processed by one or more matrix operations associated with the selection model. One or more encoder blocks and / or one or more decoder blocks may be configured to map the numerical representation of the selection task instruction to the contextual selection task instruction, particularly by considering a sequence and / or string data of one or more elements associated with the selection task instruction. The contextual selection task instruction may be associated with structured numerical data (particularly tensors), and particularly includes such structured numerical data. The contextual selection task instruction may represent a sequence of elements associated with the selection task instruction and the relationships between the elements in the sequence. The relationships between the elements can be obtained by applying one or more matrix operations to the numerical representation of the selection task instruction. Elements may include at least a portion of words, numbers, symbols, etc. Thus, the selection model can understand the relationships between elements (e.g., words in text). This improves the mapping between the selection task instruction and the model output data. Therefore, the operation engine can be operated more efficiently to process requests.

[0086] Further, the selection model may include one or more encoder outputs, particularly where the selection model may include one or more encoder blocks. Further, the selection model may include one or more decoder outputs, particularly where the selection model may include one or more decoder blocks. Additionally or alternatively, one or more encoder outputs and / or one or more decoder outputs may be configured to map context selection task instructions to multiple confidence scores associated with multiple elements (particularly a distribution of confidence scores associated with multiple elements). At least a portion of the multiple elements may be associated with, particularly included in, the model output data. One or more encoder blocks and / or decoder blocks may generate a distribution of confidence scores associated with the multiple elements (including one or more elements associated with the model output data). The model output data can be determined by selecting one or more elements associated with the model output data based on one or more confidence scores associated with one or more elements. In embodiments, one or more encoder outputs and / or one or more decoder outputs may be configured to select one or more elements associated with the model output data based on one or more confidence scores associated with one or more elements. Selecting one or more elements associated with model output data based on one or more confidence scores associated with one or more elements may include receiving a range of confidence scores and selecting one or more elements associated with one or more confidence scores within that range. One or more encoder blocks and / or one or more encoder outputs and / or one or more decoder outputs may map the numerical representation of the selection task instruction to the model output data, preferably to a distribution of confidence scores associated with multiple elements including one or more elements associated with the model output data. One or more decoder blocks and one or more decoder outputs may map the numerical representation of the selection task instruction to the model output data, preferably to a distribution of confidence scores associated with multiple elements including one or more elements associated with the operational input data.

[0087] The numerical representation of the model output data can represent the model output data (particularly a sequence and / or string data of multiple elements associated with the model output data). The numerical representation of the model output data can include the numerical representation of the model output data. The numerical representation of the selection task instruction can indicate and / or depend on a sequence and / or string data of one or more elements associated with the selection task instruction. The numerical representation of the selection task instruction can be related to at least a portion of the request and / or at least a portion of the functional specification data, and / or can represent at least a portion of the request and / or at least a portion of the functional specification data. The numerical representation of the selection task instruction can be obtained by passing the selection task instruction through one or more embedding layers. Therefore, one or more selection models can include one or more embedding layers. One or more embedding layers can be configured to map unstructured data to structured numerical representations (particularly vectorized data). The numerical representation of the selection task instruction can indicate and / or depend on a sequence and / or string data of one or more elements associated with the selection task instruction. The sequence can be encoded by positional encoding of the numerical representation of the selection task instruction. Positional encoding can be performed before the selection task instruction is provided to the structural model, or by processing the structural model (particularly one or more embedding layers). Alternatively, the encoder's self-attention mechanism may include relative position encoding as described in 1803.02155.pdf (arxiv.org). The numerical representation of the selection task instruction may include structured numerical data (particularly tensors) associated with the request and functional specification data. Specifically, the numerical representation of the selection task instruction may represent the selection task instruction, preferably at least a portion of the request and functional specification data. The numerical representation of the selection task instruction may be associated with a smaller amount of data than the selection task instruction itself. Furthermore, the numerical representation of the selection task instruction can be processed more quickly by the selection model, particularly through one or more matrix operations. Compared to the selection task instruction, the numerical representation of the selection task instruction may require less computational storage. The numerical representation of the selection task instruction may be a structured numerical representation of the selection task instruction that includes unstructured data. The numerical representation of the selection task instruction can be processed efficiently by the selection model and can save significant computational resources used for processing unstructured requests. Thus, chemical product data can be reliably generated by the selected operation engine.

[0088] In an embodiment, providing structural task instructions to the structural model may include mapping the structural task instructions to numerical representations of the structural task instructions. The structural model may be configured to map the numerical representations of the structural task instructions to numerical representations of operational input data, and vice versa. The numerical representations of the operational input data may be associated with a distribution of confidence scores associated with a plurality of elements, including one or more elements associated with the operational input data.

[0089] Numerical representations of structured task instructions can be obtained by passing the structured task instructions through one or more embedding layers. A structural model can include one or more embedding layers. These embedding layers can be configured to map unstructured data to structured numerical representations (particularly vectorized data). The numerical representations of structured task instructions can indicate and / or depend on sequences and / or strings of data that are dependent on one or more elements associated with the structured task instructions. Sequences can be encoded by positional encoding of the numerical representations of the structured task instructions. Positional encoding can be performed before the structured task instructions are fed to the structural model, or by processing the structural model (particularly one or more embedding layers). Alternatively, the encoder's self-attention mechanism can include relative positional encoding as described in 1803.02155.pdf (arxiv.org). In particular, the numerical representations of task instructions can represent structured task instructions. The numerical representations of structured task instructions can be associated with a smaller amount of data than the structured task instructions themselves. Furthermore, the numerical representations of structured task instructions can be processed by one or more matrix operations associated with the structural model. Therefore, the numerical representations of structured task instructions can be processed faster under the matrix operations of the structural model. The numerical representation of a structured task instruction can be a structured numerical representation of the structured task instruction that includes unstructured data (particularly unstructured data associated with a machine-processable numerical format, preferably a floating-point format). The structured numerical representation of the task instruction can be efficiently processed by the structured model and can save significant computational resources used for processing unstructured requests.

[0090] In an embodiment, the structural model may include one or more encoder blocks and / or one or more decoder blocks for mapping numerical representations of structural task instructions to contextual structural task instructions. Contextual structural task instructions may be associated with numerical representations of structural task instructions. Contextual structural task instructions may be numerical representations of structural task instructions obtained by processing the numerical representations of structural task instructions through one or more matrix operations associated with the structural model. One or more encoder blocks and / or one or more decoder blocks may be configured to map numerical representations of structural task instructions to contextual structural task instructions, particularly by considering sequences and / or string data of one or more elements associated with the task instruction. Contextual structural task instructions may be associated with structured numerical data (particularly tensors), and particularly include such structured numerical data. Contextual structural task instructions may represent sequences of elements associated with the structural task instruction and relationships between elements in the sequence. Relationships between elements can be obtained by applying one or more matrix operations to the numerical representation of the structural task instruction. Elements may include at least a portion of words, numbers, symbols, etc. Thus, the structural model can understand relationships between elements (e.g., words in text). This improves the mapping between structural task instructions and operational input data. Therefore, the operational engine can be operated on more efficiently to process requests.

[0091] Furthermore, the structural model may include one or more encoder outputs, particularly where the structural model may include one or more encoder blocks. Furthermore, the structural model may include one or more decoder outputs, particularly where the structural model may include one or more decoder blocks. Alternatively or additionally, one or more encoder outputs and / or one or more decoder outputs may be configured to map context structure task instructions to multiple confidence scores associated with multiple elements (particularly a distribution of confidence scores associated with multiple elements). At least a portion of the multiple elements may be associated with, and particularly included in, operational input data. One or more encoder blocks and / or decoder blocks may generate a distribution of confidence scores associated with the multiple elements (including one or more elements associated with the operational input data). The operational input data can be determined by selecting one or more elements associated with the operational input data based on one or more confidence scores associated with one or more elements. In embodiments, one or more encoder outputs and / or one or more decoder outputs may be configured to select one or more elements associated with the operational input data based on one or more confidence scores associated with one or more elements. Selecting one or more elements associated with operational input data based on one or more confidence scores associated with one or more elements may include receiving a range of confidence scores and selecting one or more elements associated with one or more confidence scores within that range. One or more encoder blocks and / or one or more encoder outputs and / or one or more decoder outputs may map the numerical representation of the structured task instruction to the operational input data, preferably to a distribution of confidence scores associated with multiple elements including one or more elements associated with the operational input data. One or more decoder blocks and one or more decoder outputs may map the numerical representation of the structured task instruction to the operational input data, preferably to a distribution of confidence scores associated with multiple elements including one or more elements associated with the operational input data.

[0092] In embodiments, any of these methods may further include generating a selection task instruction by combining request and functional specification data and / or generating a structure task instruction by merging one or more input data structures and model output data. The structure task instruction may further be associated with instructions for triggering the structure model and / or one or more data-driven models to generate operational input data. By combining the provided and / or generated data, available context can be provided to the selection model and / or structure model to accurately select at least one operational engine and / or structure the operational input data. This allows for the reliable generation of chemical product data.

[0093] In embodiments, any of these methods may further include providing a verification task instruction related to operational input data, at least one selected operational engine, and one or more input data structures (particularly at least one input data structure associated with at least one selected operational engine) to a verification model to verify the data structure associated with the operational input data. The verification model may be configured to classify whether the data structure associated with the operational input data corresponds to an input data structure associated with at least one selected operational engine. The verification model may be configured to provide an indication of whether the data structure associated with the operational input data corresponds to an input data structure associated with at least one selected operational engine. The indication may be a category label indicating whether the data structure associated with the operational input data can be verified. Therefore, the verification model may be a classification model. The classification model may be trained to provide an indication in response to receiving a verification task instruction. The verification model may include one or more classification layers. The one or more classification layers may be configured to determine, based on the verification task instruction (particularly a numerical representation of the verification task instruction), an indication of whether the data structure associated with the operational input data corresponds to an input data structure associated with at least one selected operational engine. Therefore, the indication of whether the data structure associated with the operational input data corresponds to an input data structure associated with at least one selected operational engine may be obtained from and / or may depend on the verification task instruction. Validating the data structures associated with the operational input data ensures robust triggering of at least one selected operational engine based on the operational input data. This allows for the reliable generation of chemical product data. Using a task-specific model to validate the data structures associated with the operational input data enables high validation accuracy.

[0094] In an embodiment, providing verification task instructions may include generating verification task instructions by merging one or more input data structures, at least a portion of model output data, and operational input data, and providing verification task instruction data to a verification model and / or one or more data-driven models. Verification task instructions may include one or more input data structures, instructions regarding a selected operational engine, and operational input data. Further, verification task instructions may include instructions for triggering the verification model and / or one or more data-driven models to verify data structures associated with the operational input data. By combining the provided and / or generated data, available context can be provided to the verification model and / or one or more data-driven models. This allows for accurate verification of the operational input data. This ensures robust triggering of at least one selected operational engine via the operational input data. Thus, chemical product data can be reliably generated.

[0095] In an embodiment, providing verification task instructions may include mapping verification task instructions to numerical representations of verification task instructions. The verification model may be configured to map the numerical representations of verification task instructions to numerical representations indicating whether a data structure associated with the operational input data corresponds to an input data structure associated with at least one selected operational engine. The verification model may be further configured to map the numerical representations of the indications to indications regarding whether a data structure associated with the operational input data corresponds to an input data structure associated with at least one selected operational engine.

[0096] Numerical representations of validation task instructions can be obtained by passing the validation task instructions through one or more embedding layers. A validation model may include one or more embedding layers. These embedding layers can be configured to map unstructured data to structured numerical representations (specifically, numerical representations of the data). The numerical representations of validation task instructions may indicate and / or depend on sequences and / or strings of data that are associated with one or more elements of the validation task instructions. Sequences can be encoded by positional encoding of the numerical representations of the validation task instructions. Positional encoding can be performed before the validation task instructions are provided to the validation model, or by processing the validation model (specifically, one or more embedding layers). Alternatively, the encoder's self-attention mechanism may include relative positional encoding as described in 1803.02155.pdf (arxiv.org). Numerical representations of validation task instructions may be associated with structured numerical data, particularly including such structured numerical data. Specifically, the numerical representations of validation task instructions may represent the validation task instructions themselves. The numerical representations of validation task instructions may be associated with a smaller amount of data than the validation task instructions. Furthermore, the numerical representations of validation task instructions can be processed by one or more matrix operations associated with the validation model. Therefore, the numerical representation of verification task instructions can be processed more quickly using matrix operations in the verification model. The numerical representation of verification task instructions can be a structured numerical representation of the instructions, including unstructured data (particularly unstructured data associated with machine-processable numerical formats, preferably floating-point formats). This structured numerical representation of verification task instructions can be processed efficiently by the verification model and can save significant computational resources previously used for processing unstructured requests.

[0097] The verification model may include one or more encoder blocks and / or one or more decoder blocks for mapping numerical representations of verification task instructions to contextual verification task instructions. Contextual verification task instructions may be associated with numerical representations of verification task instructions. Contextual verification task instructions can be obtained by processing the numerical representations of verification task instructions through one or more matrix operations associated with the verification model. One or more encoder blocks and / or one or more decoder blocks may be configured to map numerical representations of verification task instructions to contextual verification task instructions, particularly by considering sequences and / or string data of one or more elements associated with the verification task instructions. Contextual verification task instructions may be associated with structured numerical data (particularly tensors), and particularly include such structured numerical data. Contextual verification task instructions may represent sequences of elements associated with verification task instructions and relationships between elements in the sequence. Relationships between elements can be obtained by applying one or more matrix operations to the numerical representation of the verification task instructions. Elements may include at least a portion of words, numbers, symbols, etc. Thus, one or more data-driven models can understand relationships between elements (e.g., words in text). This improves the mapping between verification task instructions and indications regarding whether a data structure associated with the operational input data corresponds to an input data structure associated with at least one selected operational engine. Therefore, the operation engine can be operated more efficiently to process requests.

[0098] Furthermore, the verification model may include one or more encoder outputs, particularly where the verification model may include one or more encoder blocks. Furthermore, the verification model may include one or more decoder outputs, particularly where one or more data-driven models may include one or more decoder blocks. Alternatively or additionally, one or more encoder outputs and / or one or more decoder outputs may be configured to map a context verification task instruction to multiple confidence scores associated with multiple elements (particularly a distribution of confidence scores associated with multiple elements). At least a portion of the multiple elements may be associated with the instruction, particularly included in the instruction. One or more encoder blocks and / or decoder blocks may generate a distribution of confidence scores associated with multiple elements, including one or more elements associated with the instruction. The instruction can be determined by selecting one or more elements associated with the instruction based on one or more confidence scores associated with one or more elements. In embodiments, one or more encoder outputs and / or one or more decoder outputs may be configured to select one or more elements associated with the instruction based on one or more confidence scores associated with one or more elements. Selecting one or more elements associated with the indication based on one or more confidence scores associated with one or more elements may include receiving a range of confidence scores and selecting one or more elements associated with one or more confidence scores within that range.

[0099] One or more encoder blocks and / or one or more encoder outputs and / or one or more decoder outputs can map the numerical representation of a verification task instruction to the indication, preferably to a distribution of confidence scores associated with multiple elements including one or more elements associated with the indication. One or more decoder blocks and one or more decoder outputs can map the numerical representation of a verification task instruction to the indication, preferably to a distribution of confidence scores associated with multiple elements including one or more elements associated with the indication. The numerical representation of the verification task instruction can indicate and / or may depend on a sequence and / or string data of one or more elements associated with the verification task instruction. The indication may include string data indicating whether a data structure associated with the operational input data corresponds to an input data structure associated with at least one selected operational engine. The numerical representation of the indication may include numerical data (particularly structured numerical data). The numerical representation of the indication can represent the indication, particularly representing a sequence associated with the indication (particularly an indication including unstructured data). One or more encoder outputs and / or one or more decoder outputs may be further configured to map the numerical representation of the indication to an indication associated with a sequence of multiple elements or including string data.

[0100] In an embodiment, at least one selected operating engine may include a database configured to provide a digital representation of the chemical structure of a chemical product having one or more target properties in response to receiving a structured query related to one or more target properties. Operating input data may include structured queries. Chemical product data may include digital representations. At least a portion of the chemical product data can be retrieved by querying the structured database. The structured database can reliably and reliably provide data. Subsequently, using the structured database as at least one operating engine allows for the reliable and reliable generation of chemical product data. Therefore, the decisions made based on this disclosure are reasonable. This enables improved human-computer interaction, thereby providing chemical products having one or more target properties.

[0101] In an embodiment, the data generation operation engine may be configured to determine a model, particularly a data-driven model and / or a physical model, of chemical product data based on provided data generation task instructions. The physical model may include one or more equations to generate the chemical product data according to the data generation task instructions. The physical model may be associated with one or more equations relating to a functional dependency between the chemical product data and / or the data generation task instructions. This functional dependency may be based on one or more mathematical equations. The one or more mathematical equations may define a functional relationship between one or more metrics associated with the chemical product data and one or more metrics associated with the data generation task instructions.

[0102] In an embodiment, at least one selected operation engine may include a database configured to be provided with operation input data and to determine a digital representation of the chemical structure of a chemical product having one or more target properties corresponding to the operation input data in such a way as follows:

[0103] Mapping operational input data to its numerical representation

[0104] Determine one or more distances between the numerical representation of the operational input data and the numerical representation of the numerical representation of the chemical structures of two or more chemical products, wherein the numerical representation of these numerical representations is obtained by mapping the numerical representation of the chemical structures of the two or more chemical products to the numerical representation of the numerical representation of these chemical products.

[0105] Numerical representations of the chemical structures of chemical products having one or more target properties are selected by determining the numerical representation of the chemical structure associated with the minimum distance. Chemical product data may include numerical representations. Embedded databases can allow retrieval of at least a portion of chemical product data independently of keywords and therefore more accurately based on the context of the request. Thus, more accurate chemical product data can be generated. The numerical representation of the chemical structure of one or more chemical products can be obtained by using one or more embedding layers according to SMILESVec or Mol2Vec.

[0106] In an embodiment, at least one selected operation engine may include one or more sub-engines configured to perform the following:

[0107] Based on operational input data and sub-engine specification data related to one or more functions of one or more sub-engines, at least one sub-engine is selected from the one or more sub-engines to provide a chemical product with one or more target properties.

[0108] And to structure the sub-engine task instructions related to the operation input data and one or more input data structures related to one or more sub-engines, in order to generate sub-engine input data for at least one selected sub-engine.

[0109] Optionally, it classifies whether the data structure associated with the sub-engine input data corresponds to the input data structure associated with at least one selected sub-engine.

[0110] In response to providing sub-engine input data, at least a portion of the chemical product data is provided, preferably operational output data.

[0111] Sub-engine input data can be obtained from and / or may depend on the operational input data. Sub-engine input data may include at least a portion of the operational input data. At least one sub-engine may provide at least a portion of chemical product data, preferably operational output data, in response to providing sub-engine input data. In embodiments, one or more sub-engines may perform at least one of the following for each sub-engine:

[0112] At least one sub-engine is selected from the one or more sub-engines based on operational input data and sub-engine specification data related to one or more functions of one or more sub-engines.

[0113] The sub-engine task instructions related to the operation input data and one or more input data structures related to one or more sub-engines are structured to generate sub-engine input data for at least one selected sub-engine.

[0114] Optionally, a classification is made based on whether the data structure associated with the sub-engine input data corresponds to the input data structure associated with at least one selected sub-engine.

[0115] In response to providing sub-engine input data, at least a portion of the chemical product data is provided, preferably operational output data.

[0116] Or a combination thereof. This offers the advantage of breaking down a task into multiple tasks. By doing so, intermediate steps become obvious. Therefore, this feature allows for reasoning about decisions made in accordance with this disclosure. This, in turn, reduces errors in processing received requests. Thus, this contributes to the robust generation of chemical product data and enables trustworthy AI.

[0117] In an embodiment, generating chemical product data in response to providing operational input data may include:

[0118] Provides a numerical representation of the chemical structure of one or more reactants.

[0119] A numerical representation that determines the chemical structure of one or more chemical products formed in one or more chemical reactions of one or more reactants.

[0120] One or more properties associated with one or more chemical products are determined by providing a digital representation of the chemical structure of one or more chemical products to a property model, wherein the property model is configured to be provided with digital representations of the chemical products and to provide one or more properties associated with those chemical products.

[0121] Chemical products with one or more target properties are selected by comparing properties associated with one or more chemical products.

[0122] Provides a digital representation of the chemical structure of the chemical product associated with the target property. Chemical product data may include and / or represent a digital representation. Further includes determining one or more formation scores associated with the formation of one or more chemical products in one or more chemical reactions of one or more reactants. Formation scores can be determined by providing a scoring data-driven model with digital representations of the chemical structures of one or more reactants and / or one or more chemical products. The scoring data-driven model can be configured to provide one or more formation scores in response to providing digital representations of the chemical structures of one or more reactants and / or one or more chemical products. Chemical products having one or more target properties can be further selected based on one or more formation scores. One or more formation scores can be compared with a predefined range associated with the formation of chemical products having one or more target properties. Formation scores allow for assessment of the efficiency of producing one or more chemical products through one or more chemical reactions. Therefore, considering formation scores when selecting chemical products with one or more target properties allows for the selection of chemical products that can be synthesized with low resource input.

[0123] In embodiments, chemical product data may further include unstructured data. Generating chemical product data may further include providing processing task instructions related to the output of the processing model (particularly a digital representation of the chemical structure of one or more products, preferably chemical products) and a request to the processing model. The processing model may be configured to generate unstructured data in response to data provided that includes at least partially structured data. Processing task instructions can be obtained by merging the digital representation and the request. Processing task instructions may include instructions for triggering the processing model to generate chemical product data based on the operational output data and / or the digital representation of the chemical structure of one or more chemical products and the request. Further, processing task instructions may include at least a portion of the operational output data (particularly a digital representation of the chemical structure of one or more products) and the request. Chemical product data may be provided via a user interface. Alternatively or additionally, processing task instructions may be provided to one or more data-driven models. One or more data-driven models may be further configured to generate unstructured data in response to data provided that includes at least partially structured data. Providing chemical product data including unstructured data allows the chemical product data to be presented in a human-interpretable format. Therefore, human-computer interaction can be improved and trustworthy AI can be achieved by allowing the use of domain expert knowledge to verify the generated chemical product data.

[0124] In an embodiment, providing processing task instructions may include merging request and model output data and / or digital representations of the chemical structures of chemical products having one or more target properties, and providing the processing task instructions to a processing model. Merging may refer to merging request and model output data and / or digital representations of the chemical structures of chemical products having one or more target properties. Merging the request and model output data and / or digital representations of the chemical structures can produce processing task instructions that include digital representations of the request and model output data and / or the chemical structures. By combining the provided and / or generated data, available context can be provided to the processing model and / or one or more data-driven models. Thus, chemical product data can be reliably generated.

[0125] In embodiments, processing task instructions may be associated with at least a portion of chemical product data and a request. At least a portion of the chemical product data may be operational output data, i.e., as provided by at least one operational engine. The chemical product data may include operational output data and / or may indicate operational output data. If at least a portion of the chemical product data (particularly operational output data) is associated with providing a target chemical product, then the chemical product data may correspond to the request. Specifically, if providing at least a portion of the chemical product data (particularly operational output data) to the control and / or monitoring engine of the chemical production facility can trigger the production of the target chemical product and / or trigger the monitoring of the processing and / or production of the target chemical product, then the chemical product data may correspond to the request. Specifically, if at least a portion of the chemical product data (particularly operational output data) can be configured to trigger the control engine of the chemical production facility to produce the target chemical product and / or trigger the monitoring engine of the chemical production facility to monitor the processing and / or production of the target chemical product, then the chemical product data may correspond to the request.

[0126] In embodiments, a processing model can be configured to provide and / or generate chemical product data in response to determining that at least a portion of the chemical product data (particularly operational output data) corresponds to the request. Alternatively, the processing model can be configured to provide processing data related to the request, particularly indicating at least a portion of the request independent of the chemical product data. The processing data, request, and functional specification data can be provided to a selection model to evaluate the selection of at least one operational engine. The selection model can select at least one operational engine that differs from a previously selected operational engine. Operational input data associated with at least one operational engine that differs from a previously selected operational engine can be generated and / or provided to at least one operational engine. By introducing a correction loop, potential errors can be directly identified and remedied. Therefore, the accuracy of the provided chemical product data is improved. Ultimately, this contributes to improving the efficiency of monitoring and / or controlling the production and / or processing of chemical products.

[0127] In an embodiment, providing processing task instructions may include merging request and model output data and / or digital representations of the chemical structures of chemical products having one or more target properties, and providing the processing task instructions to a processing model. Merging may refer to merging request and model output data and / or digital representations of the chemical structures of chemical products having one or more target properties. Merging the request and model output data and / or digital representations of the chemical structures can produce processing task instructions that include digital representations of the request and model output data and / or the chemical structures. By combining the provided and / or generated data, available context can be provided to the processing model and / or one or more data-driven models. Thus, chemical product data can be reliably generated.

[0128] In embodiments, providing processing task instructions may include mapping processing task instructions to numerical representations of processing task instructions. A processing model and / or one or more data-driven models may be configured to map the numerical representations of processing task instructions to numerical representations of chemical product data, and / or to map numerical representations of chemical product data to chemical product data. The numerical representations of processing task instructions may be associated with a smaller amount of data than the processing task instructions. Further, the numerical representations of processing task instructions can be processed faster by the processing model, particularly through one or more matrix operations associated with the processing model. Compared to the processing task instructions, the numerical representations of processing task instructions may require less computational storage. The numerical representations of processing task instructions may be structured numerical representations of the processing task instructions, including unstructured data. The numerical representations of processing task instructions can be processed efficiently by the processing model and can save significant computational resources used for processing unstructured requests. Thus, chemical product data can be reliably generated. The numerical representations of chemical product data may be associated with a distribution of confidence scores, preferably including such distributions, of multiple elements, including one or more elements associated with the operational input data. Chemical product data can be determined by selecting one or more elements associated with the chemical product data based on one or more confidence scores associated with one or more elements.

[0129] In embodiments, any of these methods may further include, for example, mapping the numerical representation of the operational input data to the operational input data using a predefined relationship between the numerical representation of the operational input data and the operational input data. The predefined relationship may involve a vocabulary specifying the relationship between the numerical representation of the operational input data and the operational input data.

[0130] In this embodiment, providing task instructions may include at least one of the following:

[0131] The instruction for representing unstructured data, which is related to the request, the one or more input data structures, and the functional specification, is provided to one or more data-driven models.

[0132] The representation operation input data is provided to the at least one selected representation operation engine to provide a digital representation of the chemical structure of the target chemical product.

[0133] The data generation task instructions related to the digital representation of the chemical structure of the target chemical product, the request, and the functional specification data are provided to the one or more data-driven models.

[0134] The data generation operation input data is provided to the selected data generation operation engine to provide at least a portion or combination of the chemical product data. One or more data-driven models can be configured to generate representation operation input data for one or more selected representation operation engines in response to provided instructions regarding the target chemical product. The representation operation input data includes structured data used to trigger the selected representation operation engine to provide a digital representation of the chemical structure of the target chemical product. One or more data-driven models can be further configured to generate data generation operation input data for one or more selected data generation operation engines in response to provided instructions regarding the target chemical product. The data generation operation input data may include structured data (specifically, a digital representation of the chemical structure of the target chemical product) used to trigger the selected data generation operation engine to provide at least a portion of the chemical product data. Providing the chemical product data may include providing at least partially structured data. At least partially structured data can be retrieved by providing an explicit digital representation of the target chemical product. Dividing the processing of a request into subtasks allows for greater specificity of task instructions. Typically, the performance of data-driven models, such as large language models, improves with greater specificity of task instructions. Therefore, the accuracy of the generated data is improved by dividing the task of retrieving chemical product data from indications about a target chemical product into a task of providing a structured digital representation of the chemical structure of the target chemical product and a task of providing at least partially structured chemical product data based on the digital representation of the chemical structure of the target chemical product.

[0135] In an embodiment, the representation operation engine may be a database configured to provide a digital representation of a target chemical product in response to a structured query relating to an indication of the target chemical product. The representation operation input data may include the structured query. The data generation operation engine may be a database configured to provide at least a portion of chemical product data in response to a structured query relating to an indication of the target chemical product. The data generation operation input data may include the structured query.

[0136] In an embodiment, one or more data-driven models may include a representation data-driven model configured to generate representation operation input data for one or more selected representation operation engines in response to provided instructions regarding a target chemical product. The one or more data-driven models may further include a data generation data-driven model configured to generate data generation operation input data for one or more selected data generation operation engines in response to provided instructions regarding the target chemical product. Representation task instructions may be provided to the representation data-driven model. Data generation task instructions may be provided to the data generation data-driven model.

[0137] In an embodiment, one or more data-driven models may include pre-trained data-driven models. The pre-trained data-driven models can be configured to perform multiple different tasks based on multiple different task instructions. In an embodiment, this indicates that the data-driven model and / or the data-generating data-driven model can be a pre-trained data-driven model. The pre-trained data-driven model can be configured to perform multiple different tasks based on multiple different task instructions. The multiple different tasks may include structuring a request based on one or more input data structures. The multiple different task instructions may include structured task instructions. The pre-trained data-driven model may be parameterized and / or trained based on unstructured data, particularly text data, and optional numerical data (such as tabular or image data). The pre-trained data-driven model can be configured to perform multiple tasks. The pre-trained data-driven model can be configured to perform tasks based on provided task instructions. Therefore, the pre-trained data-driven model can be configured to be provided with multiple different task instructions and / or to provide multiple different types of output data when different task instructions are received.

[0138] In an embodiment, one or more data-driven models may include fine-tuned data-driven models. A fine-tuned data-driven model can be obtained by further training a pre-trained data-driven model based on training data including task instructions and corresponding operation input data. The pre-trained data-driven model can be configured to perform multiple different tasks according to multiple different task instructions. In an embodiment, the representation data-driven model and / or the data generation data-driven model can be fine-tuned data-driven models. When the representation data-driven model is a fine-tuned data-driven model, the fine-tuned data-driven model can be trained based on representation task instructions and corresponding representation operation input data. When the data generation data-driven model is a fine-tuned data-driven model, the fine-tuned data-driven model can be trained based on data generation task instructions and corresponding data generation operation input data. A fine-tuned data-driven model can be obtained by training a pre-trained data-driven model configured to perform multiple tasks according to multiple task instructions. Multiple fine-tuned data-driven models can be further trained on a training dataset including multiple task instructions of one type and corresponding output data. A fine-tuned data-driven model can be further trained to provide output data of a predefined type based on the training dataset. The fine-tuned data-driven model can be configured to provide multiple different task instructions and / or to provide multiple different types of output data when receiving different types of task instructions. Furthermore, the fine-tuned data-driven model can be configured to provide one type of output data with higher accuracy when receiving one type of task instruction than to provide other types of output data when receiving other types of task instructions.

[0139] In an embodiment, one or more data-driven models may be further configured to map numerical representations of task instructions to context task instructions and to map context selection task instructions to numerical representations of operational input data. Context task instructions can be obtained by processing the numerical representations of task instructions through one or more matrix operations associated with one or more data-driven models. The numerical representations of task instructions can also be obtained by processing context task instructions through one or more matrix operations associated with one or more data-driven models. In an embodiment, a context structure task instruction may be a numerical representation associated with the numerical representation of a structure task instruction and a relationship between two or more elements associated with the structure task instruction.

[0140] In this embodiment, unstructured data may include data generated independently of a predefined data schema and / or data format. Unstructured data may include text data, numerical data, tabular data, etc. An example of a data schema and / or data format may be JSON.

[0141] In an embodiment, the input data structure may indicate a sequence of one or more data points associated with a request. The input data structure may specify and / or define a sequence of one or more data points associated with a request. For example, the input data structure may include historical operational input data associated with one or more operating engines (particularly a selected operating engine). Alternatively or concurrently, the input data structure may indicate a pattern of one or more data points associated with a request. Operational input data may include a sequence of one or more data points associated with a request.

[0142] In embodiments, generating one or more numerical representations of the data includes providing data separated by data type to one or more embedding models. Embedding models can be configured to map data of each data type to one or more numerical representations. Embedding models for each data type can be configured to generate numerical data from non-numerical data by mapping non-numerical data to a multi-dimensional vector space. Embedding models for each data type can be configured to vectorize non-numerical data, such as text data. Embedding models for each data type can be configured to vectorize non-numerical data, such as text data and / or image data. Embedding models can be configured to map one or more data types to one or more numerical representations. Embedding models can be configured to generate joint or shared representations of one or more data types (such as text data, numerical data, and / or image data). Embedding models can be configured to generate one or more numerical representations by including mappings between elements of the data they transform. For text, such relevance can include semantics embedded in a trained probability distribution of the embedding model. Attached Figure Description

[0143] The disclosure will be further described below with reference to the accompanying drawings. In the drawings and the disclosure, the same reference numerals are intended to refer to the same or similar elements, components and / or portions.

[0144] Figure 1 An embodiment of the operating system 102 for a chemical production facility is shown.

[0145] Figure 2 Examples of methods for obtaining chemical products with target properties are shown.

[0146] Figure 3 Examples of methods for obtaining chemical product data associated with chemical products are shown.

[0147] Figure 4 Examples of methods for obtaining digital representations of chemical products are shown.

[0148] Figure 5 An embodiment of the operating system 538 is shown.

[0149] Figure 6A An embodiment of the operation engine and execution service 108 is shown.

[0150] Figure 6B An embodiment of the operation engine and execution service 108 is shown.

[0151] Figure 6C An embodiment of the operation engine and execution service 108 is shown.

[0152] Figure 6D An embodiment of the operation engine and execution service 108 is shown.

[0153] Figure 6E An embodiment of the operation engine and execution service 108 is shown.

[0154] Figure 6F An embodiment of the operation engine and execution service 108 is shown.

[0155] Figure 7 Examples of producing and / or processing chemical product 714 are shown.

[0156] Figure 8 Examples of producing and / or processing chemical product 714 are shown.

[0157] Figure 9 Examples of input and output data associated with a selection model, a structural model, a validation model, one or more operation engines, and / or a processing model are shown.

[0158] Figure 10 An example of input and output data associated with the selection model is shown.

[0159] Figure 11 Examples of input and output data associated with the structural model are shown.

[0160] Figure 12 An example of input and output data associated with the validation model is shown.

[0161] Figure 13 An example of input and output data associated with the processing model is shown.

[0162] Figure 14 An embodiment of a user interface for receiving chemical products with target properties is shown.

[0163] Figure 15 An embodiment of a user interface 1512 for receiving chemical product data is shown.

[0164] Figure 16An embodiment of a user interface 1612 for receiving a digital representation of a chemical product is shown.

[0165] Figure 17 Examples of APIs for obtaining chemical products with target properties, chemical product data associated with the chemical products, and / or digital representations of the chemical products are shown.

[0166] Figure 18 An example of training the embedding layer is shown.

[0167] Figure 19A An example of a transformer encoder architecture is shown.

[0168] Figure 19B An example of a transformer decoder architecture is shown.

[0169] Figure 19C An example of a transformer encoder-decoder architecture is shown.

[0170] Figure 20 Examples of training and / or deploying transformer encoders, transformer decoders, and / or transformer encoder-decoders are shown.

[0171] Figure 21 An example of input embedding is shown.

[0172] Figure 22 An example of input embedding is shown. Detailed Implementation

[0173] The following embodiments are merely examples for implementing the methods, systems, or application devices disclosed herein and should not be considered limiting.

[0174] Figure 1 An embodiment of an operating system 102 for a chemical production facility is shown, which is configured to provide chemical products based on one or more requests.

[0175] Chemical products are the starting materials for a variety of different end products. Therefore, chemical products must offer a range of varying properties tailored to the desired end product. The production of chemical products begins with raw materials, which are processed through one or more processing steps, including, for example, chemical reactions in a reactor and purification steps. Typically, chemical products are obtained from two or more chemical reactions that alter the chemical structure of the reactants and thus the properties of the chemical product. For example, liquids (such as monoethylene glycol) and solids (such as terephthalic acid) can be converted to produce polyesters. Polyesters are functional polymers with different properties depending on the reactants and reaction conditions. This can result in polyesters with different properties. Therefore, the chemical reactions used to produce polyesters need to be tailored to the desired properties of the polyester. Polyesters can be used in numerous fields, such as clothing or packaging. Therefore, the challenge lies in providing thousands of different chemical products to hundreds of customers, obtained from chemical production networks with multiple production steps to achieve different and customized properties. The properties of chemical products are highly dependent on their chemical structure. Even a small change in the orientation of a subgroup in a molecule with hundreds of atoms can lead to different chemical properties. Therefore, the relationship between the chemical structure and properties of chemical products is complex and difficult to control.

[0176] Subsequently, it is crucial to ensure optimal production conditions to produce chemical products with the desired properties, thereby ensuring the robust functionality of the chemical products tailored for the corresponding applications. To tailor chemical products in this highly diverse environment, obtaining digital representations of chemical structures and chemical product data is essential, indicating, for example, structural-property relationships, production conditions, and other relationships that influence the properties of the chemical product.

[0177] However, generating digital representations and chemical product data to deliver customized chemical products in chemical production environments with thousands of products and associated production lines is challenging. Multiple representations of chemical structures and multiple chemical production lines containing different equipment need to be linked sequentially according to the characteristics of the chemical products and the target chemical product.

[0178] This disclosure enables the reliable and efficient production of chemical products by processing multiple requests at different stages of the link. These requests may specifically include...

[0179] One or more requests for receiving chemical products with target properties.

[0180] One or more requests for receiving digital representations of chemical products.

[0181] One or more requests for receiving chemical product data

[0182] Or a combination thereof. Specifically, these requests can be provided independently of the predefined data structures required by the operating engine (such as a database or a data-driven model that requires predefined data structures). Despite this unstructured nature, this disclosure still enables the acquisition of chemical products, chemical product data, and / or digital representations of chemical products. In particular, purpose-specific operating engines provide accurate tools regarding their application objectives. Due to the high dependence of chemical product properties on reaction conditions, reactant ratios, chemical structures, etc., reliable and accurate chemical product data can be generated if an operating engine matching the request is selected. Furthermore, by using non-purpose-specific models, even based on unstructured requests, digital tools for generating production control data can be linked to equipment suitable for producing chemical products with target properties.

[0183] Chemical production facility 122 may include equipment for processing and / or producing chemical products. The operating system 102 of the chemical production facility may include a device interface 120. The equipment of the chemical production facility 122 may be controlled and / or monitored via the device interface 120. Therefore, the chemical production facility 122 may be monitored and / or controlled via the operating system 102 of the chemical production facility. The device interface may receive chemical product data from an output interface 118. The chemical product data may be associated with the chemical product to be produced and / or processed by the chemical production facility 122. Further, the chemical product data may be associated with the production and / or processing conditions associated with the production and / or processing of the chemical product. The chemical product data may be obtained by receiving at least one of the following:

[0184] Used to receive one or more requests for chemical products with target properties.

[0185] One or more requests for receiving digital representations of chemical products,

[0186] One or more requests for receiving chemical product data.

[0187] Or a combination thereof.

[0188] One or more requests for receiving a chemical product with target characteristics may be associated with one or more target characteristics of the chemical product and user instructions for receiving the chemical product with one or more target characteristics. Further, one or more requests for receiving a chemical product with target characteristics may be associated with a target type of the chemical product and / or a target application area associated with the chemical product. User instructions for receiving a chemical product with one or more target characteristics may include string data. User instructions for receiving a chemical product with one or more target characteristics may indicate a target quantity associated with the chemical product, a target quality associated with a chemical characteristic, a target delivery associated with the chemical product, etc. Examples of requests for receiving chemical products may be provided in [the following text is incomplete and requires further context]. Figure 14 Described in the context of.

[0189] One or more requests for receiving a digital representation of a chemical product may be associated with an indication of the chemical product. This indication may relate to one or more characteristics of the chemical product, one or more components of the chemical product, other digital representations of the chemical product (such as a common name or trade name), the type of the chemical product, and / or the application area associated with the chemical product. Examples of requests for receiving a digital representation of a chemical product may be found in... Figure 16 Described in the context of.

[0190] One or more requests for receiving chemical product data associated with a chemical product may be associated with an indication of the chemical product. This indication may relate to one or more characteristics of the chemical product, one or more components of the chemical product, other numerical representations of the chemical product (such as a common name or trade name), the type of the chemical product, and / or the application area associated with the chemical product. Examples of requests for receiving chemical product data may be found in... Figure 15 Described in the context of.

[0191] Ingestion interface 116 can be configured to receive at least one of the requests. Ingestion interface 116 can provide at least one request to selection service 104. Further, selection service can be configured to process operation instructions. Operation instructions can be associated with one or more operation engines. One or more operation engines can include the selected operation engine associated with the request. In particular, operation instructions can be associated with the technical specifications of one or more operation engines. The technical specifications of one or more operation engines can be associated with the data structure of the input data and / or output data from one or more operation engines, the technical purpose associated with one or more operation engines, the designations associated with one or more operation engines, the location of one or more operation engines, etc.

[0192] Service 104 can be configured to handle one or more requests and operation instructions. Service 104 can generate model output data. The model output data can be associated with the selected operation engine, and in particular, can instruct the selected operation engine. The selected operation engine can be configured to perform operations related to one or more received requests. Therefore, the model output data can be associated with at least a portion of the technical specifications of the selected operation engine. Further, the model output data can be associated with at least a portion of the request, particularly with target characteristics, and optionally further with target type, target application area, etc.

[0193] For example, when a request to receive a chemical product with target properties can be received, the selected operation engine can be configured to generate the chemical structure of the chemical product. This use case is... Figure 6F This is further described in the context of [the previous sentence]. Therefore, service selection 104 can select an operation engine suitable for processing at least a portion of the request. Service selection 104 can select an operation engine based on the received request (specifically, user instructions) and any one or a combination of target characteristics, target application area, target type, and instructions regarding chemical products. Service selection 104 can select an operation engine by determining a similarity score for each of one or more operation engines. The similarity score associated with the selected operation engine can be higher than the similarity scores associated with other operation engines. The similarity score can correspond to the distance between the numerical representation of the request and the numerical representation of the operation instruction. This can be [further described in the previous sentence]. Figure 2 The context will be described in further detail.

[0194] Selection service 104 can provide model output data to structure service 110. Further, the structure service can be configured to process data structures associated with one or more operation engines (including at least one data structure associated with the selected operation engine). The data structures can be provided by a data structure store 512. Structure service 110 can generate operation input data based on the model output data and the data structures. The operation input data can be associated with a data structure suitable for being provided to the selected operation engine. The operation input data can be associated with at least a portion of the model output data, specifically including at least a portion of the model output data. Therefore, structure service 110 can select a data structure corresponding to the selected operation engine and can structure the model output data to generate operation input data. Similar to selection service 104, the structure service can select a data structure associated with the selected operation engine based on one or more similarity scores associated with the operation engine associated with the data structure. Structure service 110 can provide operation input data in response to receiving model output data.

[0195] The structure service can provide operational input data to the verification service 106. The verification service 106 can be configured to verify the operational input data (particularly the data structures associated with the operational input data). Therefore, the verification service 106 can classify whether the operational input data is suitable for being provided to the selected operational engine. For this purpose, the verification service 106 can process the operational input data and the data structures associated with the operational engine. The verification service 106 can determine a confidence score associated with operational input data suitable for being provided to the selected operational engine. The confidence score can be determined similarly to the similarity score described above. The verification service 106 can provide the operational input data to the execution service 108 in response to verifying the operational input data. The execution service 108 can generate operational output data based on the operational input data. The operational output data can be associated with at least a portion of chemical product data (particularly a digital representation of the chemical structure of the chemical product). The operational output data can be requested by a received request. The operational output data may include structured data. The execution service 108 can provide the operational output data to the chemical data generation service 112. The chemical data generation service 112 can generate chemical product data based on the operational output data. Chemical product data can be associated with at least a portion of the operation output data. Further, chemical product data can be associated with system instructions. System instructions can be associated with user instructions. System instructions can, for example, instruct the delivery of a chemical product. Further, chemical product data can include unstructured data, such as string data. Chemical product data can be associated at least partially with a received request. Chemical data generation service 112 can provide chemical product data to output interface 118. In one embodiment, the output interface can be configured to provide chemical product data to device interface 120. In another embodiment, the output interface can be configured to provide chemical product data to the operating system 706 of the chemical product processing facility. Chemical data generation service 112, output interface 118, selection service 104, structure service 110, verification service 106, and execution service 106 can... Figure 5 It can be described in more detail in the context of the passage.

[0196] Figure 2 Examples of methods for obtaining chemical products with target properties are shown.

[0197] As in Figure 1As described in the context, associating chemical products with properties can be challenging due to their chemical nature. Therefore, finding chemical products for specific application areas often requires multiple testing iterations of potential candidates. Doing so consumes significant resources to obtain customized chemical products. Providing a request to the selection model and / or selection engine to receive model output data associated with the selected operation enables the reception of both unstructured and structured requests, while providing accurate and robust chemical product data associated with the chemical product. Furthermore, the chemical product can be customized for the target application area associated with it.

[0198] A request 202 for receiving a chemical product with target characteristics can be received. This request may include one or more target characteristics. The request may include additional specifications associated with the chemical product, such as a target type of the chemical product or a target application area associated with the chemical product. The target type of the chemical product may indicate, for example, whether the chemical product can be a polymer, organic liquid, inorganic salt, etc. The application area may indicate the conditions under which the chemical product and / or the product produced by processing the chemical product can be used. Further, the application area may specify use cases where the chemical product can be deployed. For example, the chemical product may be used to produce shoes (particularly soles). Use cases may indicate application characteristics, such as damping features. Damping features may be generated by the damping features of the chemical product used to produce shoes. Therefore, target characteristics may include one or more application characteristics. A request for receiving a chemical product with target characteristics may include unstructured data, such as string data, including one or more blocks of text and / or numerical data. The request may be provided by an entity for processing the chemical product. The entity for processing the chemical product may include one or more chemical product processing facilities 704. The request may be provided by, as in... Figure 7 and Figure 8 The output interface 716 of the operating system 706 of the chemical product processing facility described in the context of [the previous sentence] is provided. Requests can be made by, as in [the previous sentence]... Figure 1 , Figure 5 , Figure 7 and / or Figure 8 The ingestion interface 116 is received in the context described above.

[0199] Operation instructions 206 can be received in association with one or more operations performed by one or more operation engines. The operation instructions can describe one or more operations and / or one or more operation engines. The operation instructions can instruct the data inputs and data outputs of one or more operation engines. When a request for receiving a chemical product can be received, at least one selected operation of the one or more operations can be adapted to generate a chemical structure of a chemical product having target properties. Alternatively or additionally, at least one selected operation can be adapted to provide control data to the chemical production facility 702 to produce a chemical product having target properties. The selected operation can be performed by at least one selected operation engine of the one or more operation engines. In an embodiment, the selected operation can cause a request for further data to be provided to the entity that made the request for receiving the chemical product. The request for further data can be provided in response to determining that the operation providing the request for further data is the selected operation.

[0200] Operation instructions can be stored in an operation instruction store 506. The operation instruction store 506 may include a database. Therefore, operation instructions can be retrieved from the operation instruction store 506, for example, by providing a query for receiving an operation instruction to the operation instruction store 506. Preferably, the query for receiving an operation instruction can be provided to the operation instruction store 506 in response to receiving a request for receiving a chemical product having target properties.

[0201] Operation instructions may include unstructured data associated with descriptions of one or more operations and / or one or more operation engines. Optionally, operation instructions may include structured data associated with the structure of operation input data required by one or more operation engines to generate operation output data. Unstructured data may be, for example, string data, image data, and / or numerical data. Numerical data may be associated with numbers and optionally with corresponding units.

[0202] Operational instructions and requests can be provided to the selected model to generate model output data 208 associated with the selected operation and / or the selected operation engine. This may include generating model input data by combining operational instructions and requests for receiving chemical products with target properties. Subsequently, the model input data may include unstructured data as well as optionally structured data.

[0203] Model selection can be configured to receive model input data and generate model output data based on the model input data. Model selection can be performed as follows: Figure 18 and Figure 22 Parameterization and / or training are performed as described in the context of [the relevant context]. Model selection can be configured to map model input data to numerical representations of the model input data. Therefore, model selection can include [specific examples / methods] as described in [the relevant context]. Figure 18 The context describes one or more embedding layers, and / or one or more encoder inputs 1978, 1988 and / or one or more decoder inputs 1984, 1994. Further, the selection model can be configured to map the numerical representation of the model input data to the numerical representation of the model output data using one or more mathematical relations. Therefore, the selection model can include one or more mathematical relations. For example, the selection model can include one or more encoder blocks 1974, 1986 and / or one or more decoder blocks 1980, 1990 for mapping the numerical representation of the model input data to the numerical representation of the model output data. Further, the numerical representation of the model output data can be mapped to the model output data through one or more decoder outputs 1992, 1982 and / or encoder output 1976. Mapping the numerical representation of the model output data to the model output data can include applying one or more mathematical relations, preferably the inverse mathematical relation of a mathematical relation used to generate the numerical representation of the model output data from the model input data. By doing so, unstructured model input data is mapped to a structured representation suitable for processing by the selection model. Compared to the model input data, this structured representation requires fewer computational resources to process. Furthermore, this allows unstructured requests to be processed using the computational resources required for structured inputs.

[0204] Furthermore, the model output data can indicate the selected operation engine. Therefore, the model output data can be adapted to identify the selected operation engine. Thus, selecting the model allows the model to choose the selected operation engine to perform the selected operation. Where the request is for receiving a chemical product with target properties, the selected operation engine can be configured to provide a digital representation of the chemical structure associated with the chemical product.

[0205] In this embodiment, the model output data may be adapted to be received by at least one selected operation engine. In this embodiment, the operation instructions may further include data structures associated with one or more operation engines (including at least one data structure associated with a selected operation engine). Therefore, the model output data may include one or more data structures. The model output data obtained in response to providing a model input including data structures to a selected model may be adapted to be provided to the selected operation engine to perform the selected operation.

[0206] The model input data can be, for example, prompts used in large language models. The selection model can be a large language model. The selection model can accept other data types. For this purpose, the selection model can include multiple different embedding layers, such as... Figure 22 In the context of the image processing described, one or more embedding layers and / or as described in Figure 21One or more embedding layers described in the context of processing numerical data (especially tabular data).

[0207] Large language models can be used in conjunction with, for example, in Figures 19A to 19C The model architecture described in the context is related.

[0208] One or more data structures (including at least one data structure associated with a selected operation engine) 212 can be received, which are associated with one or more operation engines. The data structures associated with the operation engines can be stored in a data structure store 512. The data structure store 512 may include a database. Therefore, data structures can be retrieved from the data structure store 512, for example, by providing a query for receiving the data structure to the data structure store 512. Preferably, the query for receiving the data structure can be provided to the data structure store 512 in response to receiving model output data. The data structure can specify the data structure to be received by one or more operation engines (particularly the data structure to be received by at least one selected operation engine).

[0209] Operational input data 214 can be generated by providing one or more data structures and model output data to a structural model configured to generate operational input data based on one or more data structures and model output data. The operational input data can be associated with a data structure configured to be received by a selected operational engine to perform a selected operation. Therefore, the structural model can be configured to extract data required by at least one selected operational engine from an unstructured request for receiving a chemical product with target properties to perform at least one selected operation, specifically, to generate a chemical structure for a chemical product with target properties. This enables robust and customized production of chemical products. Ultimately, this results in the generation of high-quality chemical products, thereby reducing the consumption of materials that would otherwise be wasted due to substandard goods.

[0210] Specifically, generating operational input data may include generating structured input data by combining model output data and one or more data structures. The structured input data may be suitable for reception by a structured model. Preferably, the data structures and model output data may be received together to allow the structured model to select a data structure corresponding to the selected operational engine for the request. In doing so, structured operational input data can be generated for robustly providing chemical product data based on unstructured requests. The structured input data may be cue words (particularly cue words for large language models). The structured model may be a large language model.

[0211] Furthermore, generating operational input data may include providing structural input data to a structural model to generate operational input data associated with a data structure suitable for being provided to at least one selected operational engine.

[0212] Structural models can be as follows Figure 18 and Figure 22 Parameterization and / or training are performed as described in the context of [the relevant context]. The structural model can be configured to map structural input data to numerical representations of structural input data. Therefore, the structural model can include, as described in [the relevant context]... Figure 18 The context describes one or more embedding layers, and / or one or more encoder inputs 1978, 1988 and / or one or more decoder inputs 1984, 1994. Further, the structural model can be configured to map the numerical representation of the structural input data to the numerical representation of the operational input data using one or more mathematical relations. Therefore, the structural model can include one or more mathematical relations. For example, the structural model can include one or more encoder blocks 1974, 1986, one or more encoder outputs 1976, one or more decoder outputs 1992, and / or one or more decoder blocks 1980, 1990 for mapping the numerical representation of the structural input data to the numerical representation of the operational input data. Further, the numerical representation of the operational input data can be mapped to the operational input data by selecting one or more elements associated with the operational input data based on one or more confidence scores associated with one or more elements. Mapping the numerical representation of the operational input data to the operational input data can include applying one or more mathematical relations, preferably the inverse mathematical relation of a mathematical relation used to generate the numerical representation of the structural input data from the structural input data. By doing so, unstructured structural input data is mapped to a structured representation suitable for processing by the structural model. This structured representation requires fewer computational resources when processed. Furthermore, this allows unstructured requests to be processed with the computational resources required for structured input. The operational input data thus created can be suitable for being provided to the selected operational engine to execute the selected operation. Furthermore, the structured input data can instruct the selected operational engine. Therefore, the structural model can provide and / or generate data associated with the data structure required by the selected operational engine to execute the selected operation. By generating operational input data based on model output data, the data obtained by processing requests can be structured to obtain the data structure required by the selected operational engine. Subsequently, an operational engine requiring a fixed data structure (such as a database or a purpose-specific data-driven model) can be used to obtain chemical product data. Such an operational engine provides highly reliable and accurate data retrieval. Therefore, generating operational input data based on model output data enables accurate and robust generation of chemical product data. This saves significant computational resources and allows for the processing of multiple requests.

[0213] To ensure robust generation of operational input data, the data structure 218 associated with the operational input data can be verified. Verifying the data structure associated with the operational input data can refer to determining whether the data structure generated by the structural model corresponds to the input data of at least one selected operational engine. Verifying the data structure associated with the operational input data can include generating verification input data by combining the operational input data and the received data structure.

[0214] Validation input data can be provided to a validation model. The validation model can be a large language model similar to the selection model and / or the structure model. In one embodiment, the validation model can be the same model as the selection model and / or the structure model. This saves resources because fewer models need to be trained and run. In another embodiment, the validation model can include multiple deterministic functions for evaluating whether a data structure generated by the structure model associated with the operational input data corresponds to a data structure of the input data for at least one selected operational engine. Therefore, the validation model can compare the data structure associated with the operational input data with the data structure associated with the selected operational engine. In response to determining that the data structure associated with the operational input data corresponds to a data structure of the input data for at least one selected operational engine, the operational input data can be validated. Validating the operational input data can trigger the provision of the operational input data to at least one selected operational engine. Validating the operational input data allows for the validation of the data structure of the operational input data. This reduces errors generated when operating the selected operational engine. Therefore, computational resources are saved.

[0215] Operational input data can be provided to the selected operation engine to generate operational output 220 associated with the chemical product. Upon receiving a request to receive a chemical product with target properties, the operation engine can be configured to provide and / or generate a digital representation of the chemical structure of the chemical product. The operation engine can be configured as follows: Figures 6A to 6F Described in the context of.

[0216] Operational output data can be received from the operation engine. This operational output data, along with a request to receive a chemical product with the target characteristics, can be provided to the processing model to generate chemical product data 222, which includes at least a portion of the operational output data. The chemical product data can indicate the requested chemical product. The chemical product data can be a response to a request to receive a chemical product with the target characteristics. Therefore, the chemical product data can correspond to a request to receive a chemical product with the target characteristics.

[0217] Generating chemical product data can include generating processing task instructions by merging operational output data and a request for receiving chemical products with target properties (specifically, merging them into a single prompt). These processing task instructions can be provided to a processing model. The processing model can be configured to receive data and generate model input data. The processing model can include, for example, […]. Figure 18 The context describes one or more embedding layers and / or one or more encoder inputs 1978, 1988 and / or one or more decoder inputs 1984, 1994. The processing model (particularly the embedding layers of the processing model) can be configured to map processing task instructions to numerical representations of those instructions. Further, the processing model can be configured to map the numerical representations of the processing task instructions to numerical representations of the chemical product data using one or more mathematical relations. Therefore, the processing model can include one or more mathematical relations. For example, the processing model can include one or more encoder blocks 1974, 1986 and / or one or more decoder blocks 1980, 1990 for mapping the numerical representations of the processing task instructions to numerical representations of the chemical product data. Further, the numerical representations of the model's chemical product data can be mapped to the chemical product data via one or more decoder outputs 1992, 1982 and / or encoder output 1976. Mapping the numerical representations of the chemical product data to the chemical product data can include applying one or more mathematical relations, preferably the inverse of a mathematical relation used to generate the numerical representations of the chemical product data according to the processing task instructions. In doing so, unstructured processing task instructions are mapped to structured representations suitable for processing by the processing model. These structured representations require fewer computational resources when processed. Furthermore, this allows unstructured requests to be processed with the computational resources required for structured inputs. In an embodiment, the processing model can be the same model as the selection model and / or the structure model and / or the validation model. This saves resources because fewer models need to be trained and run.

[0218] The generated chemical product data can be transmitted, for example, via a user interface and / or as shown in [the document / platform]. Figure 1 The device interface 120 described in the context provides 224. Therefore, chemical product data can be adapted to be provided to a control unit of the equipment configured to control a chemical production facility. The chemical product data may include control data adapted to control a control unit of the equipment configured to control a chemical production facility.

[0219] Figure 3 Examples of methods for obtaining chemical product data associated with chemical products are shown.

[0220] It can receive a request 302 for receiving chemical product data associated with a chemical product. This request can be made as follows: Figure 2It is received as described in the context.

[0221] The request may indicate a chemical product. The request may include a numerical representation of the chemical product and / or one or more characteristics associated with the chemical product. For example, the numerical representation of the chemical product may be a designation of the chemical product and / or one or more components of the chemical product. These designations may include, for example, common names, trade names, IUPAC names, SMILES, SMARTS, etc. Further, the numerical representation of the chemical product may be an image of the chemical product and / or a measurement obtained by analyzing the chemical product using, for example, spectroscopic methods such as infrared spectroscopy.

[0222] Similar to 206, it can receive operation instructions 306 associated with one or more operations performed by one or more operation engines. The one or more operations may include a selected operation related to the request. The one or more operation engines may include a selected operation engine configured to execute the selected operation instruction.

[0223] Operating instructions and requests for receiving chemical product data can be provided to the selection model 308, as described in the context of 208.

[0224] Similar to 212, it can receive one or more data structures (including at least one data structure associated with the selected operation engine) 312.

[0225] Similar to 214, operational input data 314 can be generated by providing one or more data structures and model output data to a structure model configured to generate operational input data based on one or more data structures and model output data. The operational input data can be associated with a data structure configured to be received by the selected operational engine to execute a selected operation.

[0226] Similar to 218, the data structure 318 associated with the operational input data can be validated by providing one or more data structures and operational input data to the validation model.

[0227] Similar to 220, operation output data 320 can be generated by providing operation input data to the selected operation engine. The operation output data can be associated with a chemical product and / or the properties of that chemical product.

[0228] Similar to 222, chemical product data 322, associated with a chemical product and / or its characteristics, can be generated by providing request and operation outputs to a processing model configured to generate chemical product data based on the request and operation output data. The chemical product data may include at least a portion of the operation output data.

[0229] Where the request for receiving chemical product data includes image data, the model for processing the data associated with the request for receiving chemical product data may include, as in Figure 22 One or more embedding layers are described in the context of [the document / framework]. In cases where the request for receiving chemical product data includes numerical data (such as tabular data), the model for processing the data associated with the request for receiving chemical product data may include, as in [the document / framework]... Figure 21 One or more embedding layers are described in the context of this. In cases where the request for receiving chemical product data includes text data, the model for processing the data associated with the request for receiving chemical product data may include, as in... Figure 18 One or more embedding layers are described in the context of [the document / framework]. The model used to process requests for receiving chemical product data can be, for example, a selection model, a structural model, a verification model, one or more models that execute implementation services 6-114, a processing model, etc.

[0230] Similar to 224, chemical product data 324 can be provided.

[0231] Figure 4 Examples of methods for obtaining digital representations of chemical products are shown.

[0232] Chemical structures and complex structures are defined. Therefore, a highly standardized nomenclature is needed to describe chemical products. The thalidomide scandal is a striking example of how the orientation of functional groups in chemical products can lead to completely different behaviors and thus different properties. Similarly, chemical production is highly dependent on the chemical properties of chemical products to provide robust and customized products for further processing into final products. Therefore, clear distinctions between chemical products are of significant technical importance for ensuring robust production and processing. Obtaining a digital representation of a chemical product by submitting a request to a selection model and / or selection engine enables an efficient and robust relationship from unstructured requests to structured data. By doing so, an accurate digital representation of the chemical product associated with the request can be obtained.

[0233] Similar to 202 and / or 302, a request 402 for receiving a digital representation of a chemical product can be received.

[0234] The request may indicate a chemical product. It may include a numerical representation of the chemical product and / or one or more properties associated with the chemical product. For example, the numerical representation of the chemical product may be a designation of the chemical product and / or one or more components of the chemical product. These designations may include, for example, common names, trade names, IUPAC names, SMILES, SMARTS, etc. Further, the numerical representation of the chemical product may be an image of the chemical product and / or a measurement result obtained by analyzing the chemical product using spectroscopic methods such as infrared spectroscopy. Further, the request may indicate the type of chemical product (e.g., polymer, inorganic salt, etc.) and / or the application area associated with the chemical product.

[0235] Similar to 206 and / or 306, an operation instruction 406 may be received associated with one or more operations performed by one or more operation engines. The one or more operations may include a selected operation associated with the request. The one or more operation engines may include a selected operation engine configured to execute the selected operation instruction.

[0236] Similar to 208 and / or 308, operating instructions and a request for receiving chemical product data can be provided to selection model 408. The selection model can determine whether the request is sufficient to identify the chemical product associated with the request to receive a digital representation of the chemical product.

[0237] Similar to 212 and / or 312, it can receive one or more data structures (including at least one data structure associated with the selected operation engine) 412 that are associated with one or more operations.

[0238] Similar to 214 and / or 314, operational input data 414 can be generated by providing one or more data structures and model output data to a structure model configured to generate operational input data based on one or more data structures and model output data. The operational input data can be associated with a data structure configured to be received by the selected operational engine to perform a selected operation.

[0239] Similar to 218 and / or 318, the data structure 418 associated with the operational input data can be validated by providing one or more data structures and operational input data to the validation model.

[0240] Similar to 220 and / or 320, operation output data 420 can be generated by providing operation input data to the selected operation engine. The operation output data can be associated with a chemical product and / or the properties of that chemical product.

[0241] Similar to 222 and / or 322, chemical product data 422, associated with a chemical product and / or the characteristics of that chemical product, can be generated by providing request and operation outputs to a processing model configured to generate chemical product data based on the request and operation output data. The chemical product data may include at least a portion of the operation output data.

[0242] Where the request for receiving chemical product data includes image data, the model for processing the data associated with the request for receiving chemical product data may include, as in Figure 22 One or more embedding layers are described in the context of [the document / framework]. In cases where the request for receiving chemical product data includes numerical data (such as tabular data), the model for processing the data associated with the request for receiving chemical product data may include, as in [the document / framework]... Figure 21 One or more embedding layers are described in the context of this. In cases where the request for receiving chemical product data includes text data, the model for processing the data associated with the request for receiving chemical product data may include, as in... Figure 18 One or more embedding layers are described in the context of [the document / framework]. The model used to process requests for receiving chemical product data can be, for example, a selection model, a structural model, a verification model, one or more models that execute implementation services 6-114, a processing model, etc.

[0243] Similar to 224 and / or 324, chemical product data 424 can be provided.

[0244] Figure 5 An embodiment of the operating system 538 is shown.

[0245] Operating system 538 may include a system for obtaining chemical products, a system for obtaining chemical product data associated with the chemical products, and / or a system for obtaining digital representations associated with the chemical products. Operating system 538 may include an intake interface 116, a selection service 104, a structure service 110, a verification service, a chemical data generation service 112, and / or an output interface 118.

[0246] The intake interface 116 may be adapted to receive requests for receiving chemical product data, requests for receiving chemical products with target characteristics, and / or requests for receiving digital representations of chemical products. The intake interface may, for example, be a user interface. The intake interface 116 may allow a user to interact with the operating system 538, for example, to obtain chemical products, obtain chemical product data associated with the chemical products, and / or obtain digital representations associated with the chemical products. The intake interface 116 may be configured to receive one or more requests according to 202, 302, and / or 402.

[0247] Operation commands can be received from the operation command store 506, such as in Figures 2 to 4 As described in the context of 206, 306 and / or 406.

[0248] One or more requests can be provided to selection service 104 (specifically selection processing engine 504). Selection service may include operation instruction store 506, selection processing engine 504, and / or selection engine 508. Selection processing engine 504 can be configured to generate model input data by combining operation instructions and requests for receiving chemical product data, requests for receiving chemical products with target properties, and / or requests for receiving digital representations of chemical products, as shown in... Figures 2 to 4 The model input data can be provided to the selection engine 508, as described in the context of [the previous sentence]. Figures 2 to 4 The context describes 208, 308, and / or 408 as generating model output data. Therefore, the selection engine can include a selection model, or can connect to a selection model interface, for example, via an API. Where the selection model can connect to the selection model interface, the selection model can be configured to provide model input data to the selection model and receive model output data from the selection model. In the example, the selection model can be hosted by an entity different from the entity associated with operating system 538.

[0249] Model output data can be provided to structure service 110. Structure service 110 may include structure processing engine 514, structure engine 510, and / or data structure store 512. Data structure store 512 may store data structures associated with one or more operations (including at least one data structure associated with the selected operation engine). Data structures can be received from data structure store 512, such as in... Figure 2 As described in the context of 212. Therefore, the data structure repository can be configured to provide data structures. The structure processing engine 514 can be configured to combine data structures and model input data, as in Figures 2 to 4 The context refers to 214, 314, and / or 414. Operational input data can be provided from the structure processing engine 514 to the structure engine to generate operational input data. The structure engine 510 can be configured to generate operational input data based on 214, 314, and / or 414. Therefore, the structure engine 510 can include a structure model as described in the context of 208, 308, and / or 408. Alternatively, the structure engine 510 can interface with a structure model. Therefore, the structure model can be configured to call APIs to the structure model. Subsequently, similar to model selection, the structure model can be configured (specifically trained and / or parameterized).

[0250] Operational input data can be specifically provided from the structure engine 510 to the verification service 106. The verification service 106 may include a verification processing engine 516, a verification engine 518, and / or a data structure store 548. Alternatively, the verification service may be connected to a data structure store 512 associated with the structure service 110. In embodiments, the structure service 110 may include a structure engine 510, a structure processing engine 514, a data structure store 512, a verification processing model, and a verification engine. The verification processing engine 516 may be configured to combine data structures and operational input data according to 218, 318, and / or 418. The data structure may be received from the data structure store 512 or the data structure store 548 according to 218, 318, and / or 418. Operational input data may be provided from the verification processing engine 516 to the verification engine 518. The verification engine may be configured to verify the operational input data according to 218, 318, and / or 418. Therefore, the validation engine 518 may include a validation model as described in the context of 218, 318, and / or 418. Alternatively, the validation engine 518 may interface with a validation model. Thus, the validation engine 518 may be configured to call APIs to the validation model. Subsequently, similar to selecting a model and / or constructing a model, the validation model may be configured (specifically, trained and / or parameterized).

[0251] Operation input data may preferably be provided to execution service 108 (specifically, the selected operation engine) in response to verification of operation input data according to 220, 320, and / or 420. Execution service 108 may include one or more operation engines, which may include the selected operation engine 550. At least one of the one or more operation engines including the selected operation engine 550 may be the selected operation engine. Examples of operation engines are provided in... Figures 6A to 6F The context is described below. Operation output data can be received from the selected operation engine, as described in the contexts of 220, 320, and / or 420.

[0252] Operational output data can be provided from the selected operational engine to the chemical data generation service 112 (specifically, the output processing engine 552). The chemical data generation service 112 may include a chemical data generation engine 522 and / or an output processing engine 552. The output processing engine 552 can receive operational output data from the selected operational engine. The output processing engine 552 can receive requests from the intake interface 116 for receiving chemical product data, requests for receiving chemical products with target characteristics, and / or requests for receiving digital representations of chemical products. The output processing engine 552 can be configured to, as shown in... Figures 2 to 4The 222, 322, and / or 422 described in the context generate processing task instructions based on at least one of the following: operational output data and a request for receiving chemical product data, a request for receiving a chemical product with target characteristics, and / or a request for receiving a digital representation of a chemical product. The processing task instructions can be provided from the output processing engine 552 to the chemical data generation engine 554. The chemical data generation engine 522 can be configured to, as in... Figures 2 to 4 The 222, 322 and / or 422 described in the context generate chemical product data according to the processing task instructions.

[0253] Chemical product data can be obtained from, for example, in Figure 1 The output interface 118 described in the context and / or according to, as in Figures 2 to 4 The 224, 324 and / or 424 provided are described in the context.

[0254] Figure 6A An embodiment of an operation engine and execution service 108 using a structured database is shown.

[0255] Executing service 108 may include, for example, in Figure 5 The context describes one or more operation engines. In the example, at least one operation engine may be a structured database 650. The structured database 650 may receive operation input data from a structure service and / or a validation service 630. The structure service 630 may correspond to, as in... Figure 1 The structure service 110 is described in the context of [the relevant context]. The verification service 630 can correspond to [the specific context]. Figure 1 The verification service 106 is described in the context of the above. The structured database 650 can be configured to receive queries from the structure service and / or verification service 630. Therefore, if the selected operation engine is the structured database 650, the operation input data can include the query. The query can be associated with a predefined data structure. Further, the structured database 650 can be configured to retrieve operation output data in response to receiving a query. The structured database 650 can include predefined relationships between one or more chemical product datasets. Retrieving operation output data can include selecting at least a portion of the chemical product dataset corresponding to the query. The query can indicate at least a portion of the chemical product dataset. In the example, the structured database 650 can be an SQL database. This ensures reliable retrieval of operation output data. The operation output data retrieved by the structured database 650 can be received by the chemical data generation service 644. The chemical data generation service 644 can correspond to, as in... Figure 1 The chemical data generation service 112 is described in the context of this document.

[0256] Figure 6BAn embodiment using an operation engine and execution service 108 with an embedded database is shown.

[0257] Executing service 108 may include, for example, in Figure 5 The context describes one or more operation engines. In the example, at least one operation engine may be an embedded database 646. The embedded database 646 may receive operation input data from the structure service and / or the verification service 630. The structure service 630 may correspond to, as in... Figure 1 The structure service 110 is described in the context of [the relevant context]. The verification service 630 can correspond to [the specific context]. Figure 1 The verification service 106 is described in the context of [the context]. Operational input data can be mapped to embedded operational input data. Embedded operational input data can include a numerical representation of the operational input data. For example, embedded operational input data can be a tensor (specifically a vector). An example of embedded operational input data could be embedded input 1814. Embedded operational input data can be obtained by passing the operational input data through one or more embedding layers 1802. Examples of embedding layers and obtaining embedding layers can be found in [the context]. Figure 18 The embedded database 646 may include multiple chemical product datasets. The representation of the chemical product datasets can be obtained similarly to the representation of the operational input data. Similarly, the representation of the chemical product datasets may be an embedded chemical product dataset. Retrieving operational output data from the embedded database 646 may include selecting at least a portion of the chemical product datasets by determining whether the distance between the embedded operational input data and the embedded chemical product data is within a predefined range. The distance between the embedded operational input data and the embedded chemical product datasets may be the Euclidean distance and / or cosine distance between the embedded operational input data and the embedded chemical product data. In embodiments, chemical product datasets associated with a distance less than the distance between the embedded operational input data and other embedded chemical product datasets may be selected. This may be advantageous because operational output data can be retrieved accurately even if the operational input data may include, for example, string data. Different words used to describe the same thing may be available. The embedded database 646 can associate different words with the same meaning. Chemical products may be associated with a variety of different nomenclatures (e.g., common names or IUPAC names). Therefore, using the embedded database 646 to obtain operational output data provides accurate retrieval, even if chemical product datasets may exist from different domains. This saves resources used for coordinating documents associated with chemical product datasets. The operational output data retrieved by the embedded database 646 can be received by the chemical data generation service 644. The chemical data generation service 644 can correspond to, for example, in... Figure 1 The chemical data generation service 112 is described in the context of this document.

[0258] Figure 6C An embodiment of an operation engine based on a data-driven model and execution service 108 is demonstrated.

[0259] Executing service 108 may include, for example, in Figure 5 The context describes one or more operation engines. In an embodiment, the operation engine may be or be based on a data-driven model 652. The data-driven model 652 may be configured to receive operation input data from a structure service and / or a verification service 630. The data-driven model may generate operation output data based on the operation input data. The operation output data may be received by a chemical data generation service 644. The chemical data generation service 644 may correspond to, as in... Figure 1 The chemical data generation service 112 is described in the context of [the previous sentence]. A data-driven model may include one or more mathematical equations associated with the relationship between operational input data and operational output data. In an embodiment, the data-driven model may be a neural network. A neural network may include multiple neurons. A neuron may describe the mathematical relationship between its input and its output. A neural network may include one or more input layers, one or more hidden layers, and / or one or more output layers. An input layer may be configured to receive operational input data. The operational input data may be associated with a data structure suitable for being received by the input layer. An input layer may include multiple neurons. Neurons in the input layer may be connected to neurons in the hidden layers. Therefore, the output of the neurons in the input layer may be provided to the neurons in the hidden layers. Neurons in the hidden layers may be connected to neurons in the output layers. Therefore, the output of the neurons in the hidden layers may be provided to the neurons in the output layers. The output layer may output operational output data. The data-driven model may be adapted to describe nonlinear relationships between data points. Therefore, using a data-driven model as an operational engine allows operational output data to be obtained even when measurement data is unavailable. This saves a significant amount of resources that would otherwise be required to obtain measurement data.

[0260] Figure 6D An embodiment based on a user interface and an operation engine for executing service 108 is shown.

[0261] Executing service 108 may include, for example, in Figure 5 The context describes one or more operation engines. In embodiments, the operation engine may be or is based on user interface 654. The user interface may be configured to receive operation output data in response to providing operation input data. The operation output data may be received by chemical data generation service 644. Chemical data generation service 644 may correspond to, as in Figure 1The chemical data generation service 112 is described in the context of [the previous sentence]. Operational input data can be provided by the structure service and / or verification service 630. The user interface can receive operational output data from the user. Therefore, the user interface 654 allows the user to interact with the operating system. This is beneficial when the user is a domain expert. Chemical production has high safety requirements and therefore requires transparent decision-making, for example, when controlling chemical production facilities 122. Using the user interface as the operating engine enables efficient human-machine interaction. Furthermore, additional information for obtaining chemical products, representations of chemical products, and / or chemical product data can be received by the user interface. Therefore, requests can be enhanced, allowing for efficient request processing.

[0262] Figure 6E An embodiment of an operation engine including multiple service components and execution service 108 is shown.

[0263] Executing service 108 may include, for example, in Figure 5 The context describes one or more operation engines. In embodiments, the operation engine may include selection service 104, structure service 110, verification service 106, and / or execution service 108. Selection service 104, structure service 110, verification service 106, and / or execution service 108 may be as follows: Figure 1 The context is described below. Selection service 104 may receive operational input data from structure service and / or verification service 630. Selection service 104 may provide the processed operational input data to structure service 110. Structure service 110 may structure the processed operational input data. The structured operational input data may be provided to verification service 106 to verify the data structure associated with the structured operational input data. The structured operational input data may be provided to execution service 108 to generate operational output data. The operational output data may be received by chemical data generation service 644. Chemical data generation service 644 may correspond to, as in... Figure 1 The chemical data generation service 112 is described in the context of [the previous sentence]. Operational input data can be provided by the structure service and / or verification service 630. This can allow for customization of the selection model through further selection. For example, the selection service 104 can select a set of operation engines. Further selection can then be beneficial for further selecting one or more operation engines from that set. This can be particularly advantageous when the received request may be insufficient to determine the selected operation engine. Additional information can be provided, for example, via [the previous sentence]. Figure 6D The user interface is received within the context described. Further selections can be made by selecting service 104 based on the request and additional information. Therefore, as... Figure 6E The execution service 108 described herein can allow for efficient request processing to save computing resources.

[0264] Figure 6F An embodiment of an operation engine is shown, which includes multiple services for determining chemical products based on reactants and execution service 108.

[0265] Executing service 108 may include, for example, in Figure 5 The context describes one or more operation engines. In embodiments, the operation engine may include an intake interface 6-124, a chemical structure generation engine 6-132, a compound database 6-126 (including digital representations of the chemical structures of one or more reactants 6-128), a property determination engine 6-134, a formation score determination engine 6-130, a product determination engine 6-136, and / or an output interface 6-138. The intake interface 6-124 may be configured to receive operation input data from a structure service and / or a verification service 630. The operation input data may be provided by the structure service and / or the verification service 630. The structure service 630 may correspond to, as in Figure 1 The structure service 110 is described in the context of [the relevant context]. The verification service 630 can correspond to [the specific context]. Figure 1 The verification service 106 is described in the context of [the previous sentence]. The ingestion interface 6-124 can provide operational input data to the chemical structure generation engine 6-132. The chemical structure generation engine 6-132 can be configured to generate digital representations of chemical products based on the operational input data. The operational input data may include target properties of the chemical products. The chemical structure generation engine 6-132 can determine the digital representation of the chemical product obtained through a chemical reaction of one or more reactants based on the digital representations 6-128 of the chemical structures of one or more reactants. For this purpose, the chemical structure generation engine 6-132 can receive digital representations 6-128 of the chemical structures of one or more reactants from the compound database 6-126. A request for receiving digital representations 6-128 of the chemical structures of one or more reactants can be provided by the chemical structure generation engine 6-132 to the compound database 6-126. The request for receiving digital representations 6-128 of the chemical structures of one or more reactants can be a query suitable for reception by the compound database 6-126. The compound database 6-126 can be as follows: Figure 6A The structured database described in the context of [the context].

[0266] A numerical representation of the chemical structure of a chemical product can be provided to a formation score determination engine 6-130. The formation score determination engine 6-130 can be configured to determine the formation score associated with the formation of the chemical product from one or more reactants. For example, a high formation score can indicate a high formation rate of the chemical product. The formation score can be obtained by calculating the extent to which the atomic configurations are not altered by the chemical reaction from one or more reactants to the chemical product. Since the production of a chemical product may be associated with equilibrium and incomplete transformation, the formation score can indicate the efficiency of the production process. If the formation score associated with the formation of the chemical product can be within a predefined range, the chemical product can be selected. This allows for increased production efficiency of the chemical product and reduced amounts of undesirable byproducts.

[0267] Furthermore, the numerical representation of the chemical structure of the chemical product can be received by the characteristic determination engine 6-134. The characteristic determination engine 6-134 can be configured to determine the characteristics of the chemical product based on the numerical representation of the chemical product. For example, the characteristic determination engine 6-134 may include a classification model.

[0268] The determined characteristics can be provided to the product determination engine 6-136. The determined formation score can be provided to the product determination engine 6-136. The digital representation of the chemical structure of the chemical product can be provided to the product determination engine 6-136. The product determination engine 6-136 can be configured to select chemical products associated with target characteristics, for example, by comparing the characteristics of the chemical products with target characteristics. The product determination engine 6-136 can further select chemical products by determining that the formation score is within a predefined range. Therefore, the characteristic determination engine 6-134 can select chemical products with target characteristics from those chemical products associated with the digital representation of the chemical structure of the chemical product obtained by the chemical structure generation engine 6-132. The product determination engine 6-136 can provide the digital representation of the chemical structure to the output interface 6-138. Therefore, the operational output data can be associated with the digital representation of the chemical structure of the chemical product associated with the target characteristic.

[0269] By using the engine described above and / or performing the actions described above, chemical products with target properties can be efficiently obtained from reactants available, for example, in chemical production facility 122.

[0270] The operation output data can be received by the chemical data generation service 644. The chemical data generation service 644 can correspond to, for example, in... Figure 1 The chemical data generation service 112 is described in the context of this document.

[0271] Figure 7 Examples of producing and / or processing chemical product 714 are shown.

[0272] Chemical products can be produced by chemical production facility 702. Chemical products can be supplied from chemical production facility 702 to chemical product processing facility 704. Chemical product processing facility 704 can be controlled and / or monitored by operating system 706 of chemical product processing facility.

[0273] Chemical production facility 702 can be connected to operating system 102 of the chemical production facility. Therefore, chemical production facility 702 can be monitored and / or controlled by operating system 102 of the chemical production facility. Operating system 102 of the chemical production facility can be as follows: Figure 1 Described in the context of.

[0274] The operating system 706 of the chemical product processing facility may include an ingestion interface 712, a service request 710, and / or an output interface 708. For processing a chemical product 714, the chemical product 714 may be associated with a target characteristic. The target characteristic may be specified by the chemical product processing facility 704 and / or may be the result of target processing of the chemical product. Therefore, a processing specification may be provided to the ingestion interface 712. The ingestion interface 712 may be configured to receive the processing specification. The ingestion interface 712 may provide the processing specification to the service request 710. The service request 710 may be configured to generate a request for receiving the chemical product associated with the target characteristic based on the processing specification. This request may be provided to the output interface 708. The output interface 708 may provide the request to the ingestion interface 716. The request may be provided as shown in... Figures 2 to 4 Process it as described in the context.

[0275] Figure 8 Examples of producing and / or processing chemical product 814 are shown.

[0276] Chemical products can be produced by chemical production facility 802. Chemical production facility 802 can be connected to operating system 102 of the chemical production facility. Therefore, chemical production facility 802 can be monitored and / or controlled by operating system 102 of the chemical production facility. Operating system 102 of the chemical production facility can be as follows: Figure 1 The chemical product can be supplied from the chemical production facility 802 to the chemical product processing facility 804. The chemical product processing facility 804 can be controlled and / or monitored by the operating system 804 of the chemical product processing facility. The operating system of the chemical product processing facility 804 may include an ingestion interface 812, a service request 810, a device interface 824, and / or an output interface 808.

[0277] The service request 810 can generate a request for receiving chemical product data, for example, to adapt the chemical product processing facility 804 based on the received chemical product and / or the characteristics of the chemical product. Improper handling of chemical products can significantly reduce their performance during processing and / or application. Therefore, it may be necessary to control the chemical product processing facility 804 based on the chemical products received from the chemical production facility 122. For this purpose, the request generated by the service request 710 can be provided to the output interface 808. The output interface 808 can provide the request to the intake interface 116 of the operating system 102 of the chemical production facility. The operating system 102 of the chemical production facility can be as follows: Figures 1 to 4 The request is processed as described in the context. The output interface 118 of the operating system 102 of the chemical production facility can provide chemical product data to the ingestion interface 812 of the operating system of the chemical product processing facility 804. Further, the chemical product data can be provided by the ingestion interface 812 to the device interface 824. The device interface 824 can be configured to control the chemical product processing facility 804. Therefore, the processing of the chemical product can be improved by retrieving the chemical product data. Advantageously, the request can include unstructured data, while the chemical product data can be structured and therefore machine-readable. By doing so, the request service 710 can include, for example, a user interface for inputting string data, while the chemical product processing facility 804 can be controlled via structured control data obtained from the chemical product data.

[0278] Figure 9 Examples of input and output data associated with a selection model, a structural model, a validation model, one or more operation engines, and / or a processing model are shown.

[0279] The selection model can be configured to receive selection task instructions that include request and functional specification data. The selection model can then generate model output data based on these instructions. Examples of input and output data associated with a selection model can be found in [link to relevant documentation]. Figure 10 As seen in [the documentation], model output data can be combined with one or more input data structures to form a structured task instruction. The structured model can be configured to generate operational input data based on the structured task instruction. Examples of input and output data associated with a structured model can be found in [the documentation]. Figure 11 As seen in [the document], operational input data can be combined with one or more input data structures and instructions regarding at least one selected operational engine to form a verification task instruction. The verification model can be configured to generate verification instructions based on the verification task instruction. Examples of input and output data associated with the verification model can be found in [the document]. Figure 12As seen in [the document], the verification indication can be an indication of whether the data structure associated with the operation input data corresponds to the input data structure associated with at least one selected operation engine. If the data structure associated with the operation input data can be verified, the operation input data can be provided to one or more selected operation engines to generate operation output data. An example of at least one selected operation engine can be found in [the document]. Figure 6F The description is within the context of [the relevant information]. Operational output data can be combined with the request to form a processing task instruction. The processing model can be configured to generate chemical product data based on the processing task instruction. Examples of input and output data associated with the processing model can be found in [the relevant information]. Figure 13 I saw it in the middle.

[0280] Figure 10 An example of input and output data associated with the selection model is shown.

[0281] Figure 11 Examples of input and output data associated with the structural model are shown.

[0282] Figure 12 An example of input and output data associated with the validation model is shown.

[0283] Figure 13 An example of input and output data associated with the processing model is shown.

[0284] Figure 14 A user interface 1402 is shown for receiving chemical products with target properties.

[0285] A request to receive a chemical product with target characteristics may include the target characteristics, the target type of the chemical product, the target application area associated with the chemical product, and a description of the requested service. In the example, the description of the requested service may include unstructured text data. The description of the requested service may be predefined and / or can be specified by the user. The target characteristics, target type, and / or target application area can be entered into the user interface, for example, by selecting a target value from multiple values, or by specifying free text associated with the target characteristics, target type, and / or target application area. For this purpose, the user interface 1402 may provide corresponding input fields 1408, 1404, 1410, and / or 1412. This can be schematically depicted in the upper user interface 1402. Once data can be entered into the user interface 1402, the user interface 1402 can display the entered data (as depicted in the lower user interface 1402 below) in the corresponding fields and provide options for producing the requested chemical product.

[0286] Figure 15 An embodiment of a user interface 1512 for receiving chemical product data is shown.

[0287] Requests for obtaining chemical product data can be received and / or provided by user interface 1512. The request may include the name of the chemical product, the type of the chemical product, the application area associated with the chemical product, and / or a description of the requested service. In the example, the description of the requested service may include unstructured text data. The description of the requested service may be predefined and / or specified by the user. The name of the chemical product, the type of the chemical product, and / or the application area of ​​the chemical product can be entered into the user interface, for example, by selecting a target value from multiple values, or by specifying free text related to the target characteristic, target type, and / or target application area. For this purpose, user interface 1502 can provide corresponding input fields 1504, 1506, 1508, and / or 1510. This can be schematically depicted in the upper user interface 1512. Once data can be entered into user interface 1512, user interface 1512 can display the entered data (as depicted in the lower user interface 1512 below) and the requested chemical product data in the corresponding fields.

[0288] Figure 16 An embodiment of a user interface 1612 for receiving a digital representation of a chemical product is shown.

[0289] Requests for obtaining chemical product data can be received and / or provided by user interface 1612. The request may include the name of the chemical product, the type of the chemical product, the application area associated with the chemical product, one or more components of the chemical product, and / or a description of the requested service. In the example, the description of the requested service may include unstructured text data. The description of the requested service may be predefined and / or can be specified by the user. The name of the chemical product, the type of the chemical product, one or more components of the chemical product, and / or the application area can be entered into the user interface, for example, by selecting a target value from multiple values, or by specifying free text related to the target characteristic, target type, one or more components, and / or target application area. For this purpose, user interface 1612 can provide corresponding input fields 1604, 1606, 1608, 1624, and / or 1610. This can be schematically depicted in the upper user interface 1612. Once data can be entered into user interface 1612, user interface 1612 can display the entered data (as depicted in the lower user interface 1612 below) and a numerical representation of the requested chemical product in the corresponding fields. In the example, the numerical representation of the requested chemical antifreeze product could be a list of ingredients. Such a list of ingredients may be necessary when processing of a particular chemical should be prevented. Even small amounts of a chemical product can have a significant impact on its processing. For example, small amounts of metal ions adhering to a stir bar, even after thorough cleaning, can catalyze unwanted reactions, thereby altering the properties of the chemical product. Therefore, accurate numerical representation of the chemical product is crucial for ensuring efficient processing.

[0290] Figure 17 Examples of APIs for obtaining chemical products with target properties, chemical product data associated with the chemical products, and / or digital representations of the chemical products are shown.

[0291] To obtain chemical products with target characteristics, associated chemical product data, and / or digital representations of the chemical products, a selection model can be deployed. The selection model can be described in more detail in 208, 308, and / or 408. The selection model can be invoked via the selection model API 1708. The selection model API 1708 can be configured to receive operational instructions and requests for obtaining chemical products with target characteristics, associated chemical product data, and / or digital representations of the chemical products (specifically, model input data). Furthermore, the selection model API 1708 can be configured to receive model output data from the selection model.

[0292] Model output data and data structures (specifically, structure input data) associated with one or more operation engines can be received by the structure model API 1710. Further, the structure model API 1710 can provide structure input data to the structure model as described in the contexts of 214, 314, and / or 414. Further, the structure model API 1710 can be configured to receive operation input data from the structure model.

[0293] Operational input data and data structures can be provided to the verification model via the verification model API 1712. Furthermore, the verification model API 1712 can be configured to receive verified operational input data from the verification model. The verification model can be described in the context of 218, 318, and / or 418.

[0294] Operation input data can be received by the operation engine API 1714 and provided to the operation engine as described in the contexts of 220, 320, and / or 420. The operation engine API 1714 can be configured to receive operation output data from the operation engine.

[0295] Operational output data and requests (specifically processing task instructions) for obtaining chemical products with target characteristics, chemical product data associated with the chemical products, and / or digital representations of the chemical products can be provided to the processing model API 1716. 1716 can be configured to receive processing task instructions and provide them to the processing model. The processing model can be described in the context of 222, 322, and / or 422. Further, 1716 can be configured to receive chemical product data from the processing model.

[0296] As in Figure 1 , Figure 7 and / or Figure 8 The device interface 120 described in the context may include device API 1718. Device API 1718 may be configured, for example, to receive chemical product data from output interface 118 and provide the chemical product data to the device of chemical production facility 122.

[0297] Figure 18An embodiment of obtaining an embedding layer is illustrated. An embedding layer can be obtained by training, for example, a Continuous Bag-of-Words (CBOW) model or a skip-gram model. The embedding layer can be adapted to generate embedded input data based on input data. Generating embedded input data can refer to embedding the input data. The embedding layer can map data to a numerical representation of the data. Embedded data can be used synonymously with the numerical representation of the data. Embedding input data can produce a representation associated with the input data. Therefore, embedded input 1814 can be a representation associated with the input data. The input data can include one or more elements. One or more elements can be represented by input vector 1806. In particular, embedded input 1814 and / or input vector 1806 can be machine-readable and / or processor-processable. For this purpose, embedded input 1814 and / or input vector 1806 can be tensors, particularly first-order tensors. Specifically, input vector 1806 can be a one-hot vector or a sum of multiple one-hot vectors. A one-hot vector can be a vector with a non-zero entry. Examples of one-hot vectors can be 1808, 1810, and 1812. Non-zero entries in the one-hot vector and / or input vector 1806 can indicate elements. For example, a lookup table can define the relationship between the position of a non-zero entry and the element indicated by the one-hot vector. The lookup table can specify multiple distinct elements. The number of distinct elements can be equal to the number of entries in the one-hot vector. The number of distinct elements can be referred to as the vocabulary. In the example, elements can be represented by tokens, and a sequence of elements can refer to at least a part of a sentence. At least a part of a sentence can be represented by multiple tokens. Tokens can represent at least a part of an element and / or a word. For example, in cases where an element will be associated with only one word, words such as “embeddings,” “embedding,” or “embed” will constitute distinct elements. The first token can represent the stem “embed,” while the suffix, which typically appears in multiple words, can be represented by the second, third, and fourth tokens. The second, third, and fourth tokens can be used to represent other words, such as “look,” “looking,” etc., preferably used in conjunction with the fifth token representing the stem “look.” Ultimately, this word segmentation of elements associated with multiple stems and multiple suffixes results in fewer lexical units used to represent multiple elements, and therefore uses fewer computational resources.

[0298] A lookup table specifying a subset of the vocabulary of a language such as English can include 10,000 words or more. The embedded input 1814 can be a lower-dimensional representation than the input vector 1806. For example, a typical embedded input 1814 can include hundreds of different entries. Therefore, the embedded input 1814 constitutes a denser representation of one or more elements using fewer computational resources. Furthermore, the embedded input 1814 can represent relationships between two or more elements. For example, the words “Italy” and “Germany” can be similar or more closely related because they both define European countries, while the word “example” may be quite different from these two corresponding words. The smaller the dot product between two embedded inputs 1814, the more similar the two elements associated with the embedded input 1814 can be. Therefore, the embedded input 1814 can accurately represent one or more elements and produce accurate results based on the processing of the embedded input 1814.

[0299] To transform input vector 1806 into embedded input 1814, the embedding layer can include a number of neurons equal to the number of entries in embedded input 1814. Based on embedded input 1814, the output layer can generate output vector 1816. The output vector can be a vector and / or can indicate one or more elements. Output vector 1816 can indicate one or more elements that are different from input vector 1806 and / or different from the one-hot vector associated with input vector 1806. For this purpose, the output layer can include a number of neurons equal to the number of entries in input vector 1806 and / or output vector 1816. The output layer can apply a softmax function to embedded input 1814. By doing so, the output vector can include probabilities associated with elements associated with non-zero entries in output vector 1816. Therefore, one or more elements with corresponding probabilities can be obtained from output vector 1816. Where input vector 1806 can specify one or more sequences of elements, output vector 1816 can specify one or more elements corresponding to the sequence of elements specified by input vector 1806. Figure 18 In the example, the element associated with vector 1818 corresponds to the input vector with a 71% probability. Additional or substitute elements can correspond to the input vector as indicated by the output vector with a lower probability. By defining a threshold that can be compared with probabilities, the selection of corresponding elements can be customized to the user's needs. Elements generated by the model, including embedding layer 1802 and output layer 1804, can refer to the most probable element indicated by output vector 1816. Therefore, Figure 18 The model described can generate elements associated with vector 1818 with a 71% confidence score.

[0300] Figure 18The model can be a Continuous Bag-of-Words (CBOW) model. A CBOW model can be trained on a training dataset that includes multiple input vectors and corresponding output vectors. Since the training dataset may be unlabeled, training the CBOW model can be referred to as self-supervised training. Before training the CBOW model, it can be initialized with random values ​​for the weights assigned to neurons. During training, the input vector can be used to initialize the embedding and output layers, and the loss can be determined by comparing the output vector obtained by passing the input vector 1806 through the model with the output vector corresponding to the input vector 1806 as specified by the training dataset. Based on the determined loss, backpropagation can be applied to determine the gradients associated with the neurons in the embedding layer 1802 and the output layer 1804 to reduce the loss. The neuron weights can be updated using a gradient descent algorithm based on the determined gradients. If the CBOW model achieves the predetermined loss, training can be terminated, and the trained CBOW model is obtained. Based on the trained CBOW model, the embedding layer 1802 can be adapted to embed input data including one or more elements. This embedding layer 1802 can be used in other machine learning architectures that require an embedding layer 1802, such as in... Figure 19A , Figure 19B and Figure 19C The context describes the transformer encoder, transformer decoder, or transformer encoder-decoder architecture. Training these architectures may require trained embedding layers 1802. Therefore, models such as the CBOW model can be trained before training the transformer encoder, transformer decoder, or transformer encoder-decoder architecture.

[0301] Figure 19A An embodiment of a transformer encoder architecture is illustrated. The transformer encoder includes an encoder input 1978, one or more encoder blocks 1974, 1914, and an encoder output. The transformer encoder architecture can be derived from, as is known in the art and as... Figure 19C The transformer encoder-decoder architecture shown is obtained. Specifically, the transformer encoder can be referred to as an X-former. The transformer encoder architecture can correspond to an encoder architecture associated with the transformer encoder-decoder architecture, but has additional encoder outputs instead of directly connecting the encoder block to the decoder of the transformer encoder-decoder architecture. Various transformer encoder architectures are available in the art, such as the bidirectional transformer encoder representation (BERT).

[0302] Input data can be received at encoder input 1978. Input embedding 1902 can be applied to encoder input 1978. Applying input embedding 1902 can instruct input data to pass through an embedding layer, for example, as in... Figure 18 Described in the context of [the above]. Further, the encoder input 1978 may apply position coding 1904. Applying position coding 1904 may refer to adding position factors to the embedded input obtained via input embedding. Preferably, the input data may specify a sequence of elements. Position factors It can indicate the position of an element within a sequence. For example, the position factor can be obtained based on the following equation. :

[0303]

[0304]

[0305] Here, pos can refer to the position of an element within the sequence, i can refer to the dimension associated with the input embedding, and d can refer to the dimension of the model (e.g., a transformer decoder, transformer encoder, or transformer encoder-decoder). This can be referred to as absolute position embedding. Alternatively, positional encoding can be based on Rotated Position Embedding (RoPE). Positional encoding is advantageous because it allows for processing sequential data without requiring further dimensions indicating the position of each element. Therefore, positional encoding 1904 reduces the computational resources required for embedding the input data. By passing the input data through the encoder input, the input data can be transformed into a second-order tensor representing the sequence of elements. This second-order tensor can be referred to as embedded input data. Embedded input data can be processed by the encoder block. Embedded input data can be provided to layer normalization 1908 via residual connections. Multi-head self-attention 1906 can be applied to the embedded input data. Multi-head self-attention 1906 can include two components: multi-head and self-attention. Self-attention can be understood as a filter applied to the embedded input data. By applying filters to embedded input data, elements associated with the embedded input data that contribute to the generated output data can be identified. Therefore, a filter can represent the degree of contribution of elements associated with the embedded input data to the generated output data. Applying filters can be described as weighting the elements associated with the embedded input data. This is particularly advantageous for long sequences of elements. Filters can be learned and improved during training by learning to identify the contributions of elements associated with the embedded input data. For example, in a partial sentence "I went to the bakery to buy a," the last word can be generated by a data-driven model such as a transformer encoder. Self-attention can cause the transformer encoder to focus primarily on the words "bakery" and "buy" to generate the word "bread." Self-attention can refer to attention generated based on input data. Therefore, filters can be determined based on the input data, preferably the embedded input data. The embedded input data can be used as the query Q, key K, and value V for the self-attention operation. Self-attention can refer to attention based on the received input data. Therefore, the filter can be calculated based on the following formula by inserting the corresponding tensor based on the embedded input data:

[0306]

[0307] in, The dimension corresponding to the key.

[0308] To further improve the efficiency of the transformer encoder, multiple heads are used to apply filters, resulting in multi-head self-attention 1906. Multi-head self-attention 1906 can include applying filters to two or more parts of the embedded input data. Therefore, the tensor can be split into two or more parts, and filters can be applied to each of the two or more parts separately through the two or more heads according to the following equation:

[0309] Where the parameter matrix Where i can refer to the number of heads. , and It can refer to the value, key, and query dimension.

[0310] The results of two or more heads can be cascaded according to the following equation:

[0311] in, And h can refer to the number of heads.

[0312] Embedded input data can be transformed into a context tensor via multi-head self-attention 1906. The context tensor can represent a sequence of elements in the input data and the relationships between two or more elements. The context tensor can be a second-order tensor and / or may include one or more first-order tensors. Following multi-head self-attention 1906, layer normalization 1908 can be applied based on the context tensor and / or the embedded input data from the residual connections. Applying layer normalization 1908 can refer to normalizing the context tensor. Normalizing the context tensor reduces the values ​​of the entries in the context tensor. This reduces the computational cost associated with processing the context tensor. Furthermore, it improves training by promoting loss convergence and preventing instability.

[0313] After layer normalization 1908, the context tensor can be passed to the feedforward layer 1910 again, followed by layer normalization 1912 based on the residual connections to the context tensor and / or the output of the feedforward layer 1910. The feedforward layer 1910 can be a feedforward neural network. The feedforward neural network can include multiple fully connected neurons. Passing the context tensor through the feedforward neural network can result in a linear transformation of the context tensor. Alternatively, the neural network can include one or more activation functions, such as rectified linear units (ReLU). Thus, the neural network can be configured to perform one or more nonlinear operations on the context tensor and / or nonlinearly transform the context tensor. After the context tensor has been transformed and / or normalized by the feedforward layer 1910 and layer normalization 1912, the context tensor can be provided to one or more additional encoder blocks 1914. Passing the context tensor through the feedforward layer 1910 can adapt the context tensor to the processing of another attention layer of one or more additional encoder blocks 1914 in order to apply a self-attention filter, preferably a multi-head self-attention 1906. The context vector, after being transformed by layer normalization 1912 and feedforward layer 1910, can be referred to as the hidden state.

[0314] The encoder output 1976 includes a linear layer 1916 and a softmax layer 1918. The linear layer 1916 transforms the context vector into a logits vector. The linear layer can be fully connected. The logits vector obtained by passing the context tensor through the linear layer 1916 can be passed through the softmax layer 1918. Passing the logits vector through the softmax layer 1918 can mean applying the softmax function to the logits vector. Applying the softmax function to the logits vector produces a probability distribution of one or more elements corresponding to the sequence of elements in the input data. The probability distribution of one or more elements can be confidence scores associated with one or more elements. Based on predefined selection criteria, one or more elements can be selected according to the probability distribution. The one or more selected elements can be referred to as one or more elements generated by the transformer encoder. One or more generated elements can be provided to the encoder input to generate another one or more elements corresponding to the sequence of input data and the one or more elements generated by the transformer encoder, as shown in... Figure 20 Described in the context of.

[0315] Figure 19B An example of a transformer decoder architecture is shown.

[0316] The transformer decoder includes a decoder input 1984, one or more decoder blocks 1980, 1932, and a decoder output 1992. The transformer decoder architecture can be derived from, as is known in the art and as... Figure 19C The transformer encoder-decoder architecture shown is obtained. The transformer decoder can be referred to as X-former. The transformer decoder architecture can correspond to the decoder architecture associated with the transformer encoder-decoder architecture, but does not depend on receiving one or more hidden states from the encoder of the transformer encoder-decoder. There are various transformer decoder architectures available in the art, such as Generative Pretrained Transformer (GPT).

[0317] The decoder input terminal 1984 can be applied to, for example, in Figure 19A The input embedding 1902 and positional encoding 1904 described in the context are similar to the input embedding 1920 and positional encoding 1922.

[0318] Decoder block 1980 may include layer normalization 1926, masked multi-head self-attention 1924, feedforward layer 1928, and / or layer normalization 1930. Embedded input data generated by passing input data through decoder input 1984 can be provided to layer normalization 1926 via residual connections. Further, masked multi-head self-attention 1924 can be applied to the embedded input data. Masked multi-head self-attention 1924 corresponds to... Figure 19A The multi-head self-attention 1906 described in the context of [previous context] further masks the portion of the embedded input data associated with elements in the sequence that are later than the elements to be generated. Alternatively, the portion of the input data associated with elements in the sequence that are later than the elements to be generated may not be received and / or transformed into embedded input data. Therefore, a transformer decoder can be adapted to generate subsequent elements of a sequence, while a transformer encoder can be adapted to generate missing elements within a sequence and / or between two or more sequences. Thus, a transformer encoder can be configured to perform a classification task. A transformer decoder can be configured to generate text.

[0319] Similar to Figure 19A The transformer encoder described in the context can generate a context tensor by applying masked multi-head self-attention 1924 and layer normalization 1926. The context tensor can be provided to layer normalization 1930 via residual connections. Further, feedforward layers 1928 and layer normalization 1930 can be similar to... Figure 19A The context tensor is described in the feedforward layer 1910 and layer normalization 1912. The context tensor can be provided to one or more additional decoder blocks 1932.

[0320] The decoder output 1992 may include a linear layer 1934 and a softmax layer 1936. The linear layer 1934 and softmax layer 1936 can be similar to... Figure 19A The linear layer 1916 and softmax layer 1918 are described in the context of this study.

[0321] Figure 19C An embodiment of a transformer encoder-decoder architecture is illustrated. The transformer encoder-decoder may include an encoder input 1988, one or more encoder blocks 1986, 1964, a decoder input 1994, a decoder block 1990, and a decoder output 1992. The encoder input 1988 may correspond to... Figure 19A The encoder input terminal 1978. One or more encoder blocks 1986, 1964 can correspond to Figure 19A One or more encoder blocks 1974, 1914. Decoder input 1994 can correspond to... Figure 19B The decoder input terminal 1984.

[0322] Decoder block 1990 may include, as in Figure 19B The masking multi-head self-attention 1924, layer normalization 1926, feedforward layer 1928, and layer normalization 1930 described in the context are similar to the masking multi-head self-attention 1970, layer normalization 1972, feedforward layer 1938, and layer normalization 1940. Decoder block 1990 may further include multi-head self-attention 1950 and layer normalization 1948. Similar to... Figure 19B The description can be derived from Masked Multi-Head Self-Attention 1970 and Layer Normalization 1972, where the context tensor can be obtained. Figure 19A Multi-head self-attention 1906, similar to multi-head self-attention 1950, is applied to the context vector obtained from layer normalization 1972 and the hidden states of one or more encoder blocks 1986, 1964. Layer normalization 1948 can be applied to the context vector obtained from multi-head self-attention 1950 and the context vector provided via residual connections from layer normalization 1972. Similar to... Figure 19BThe description suggests that the context vector generated by layer normalization 1948 can be processed via feedforward layer 1938 and layer normalization 1940. The context vector generated by layer normalization 1940 can be provided to another decoder block 1942, similar to decoder block 1990. The context vector obtained from one or more decoder blocks 1990, 1942 can be provided to decoder output 1992. Decoder output 1992 can correspond to... Figure 19B The decoder output is 1982.

[0323] Using the above architecture, the transformer encoder-decoder can receive and process input data at encoder input 1988 and one or more encoder blocks 1986, 1964, as well as decoder block 1990 and decoder output 1992. Based on the input data, the transformer encoder-decoder can generate output data partially or sequentially. Sequentially generated output data can be provided to decoder input 1994, one or more decoder blocks 1990, 1942, and decoder output 1992, and / or processed by them. Preferably, a sequence can be provided to encoder input 1988, and after at least a portion of the generated output data has been generated, at least a portion of the elements of the generated output data can be provided to decoder input 1994. By doing so, by considering the input data and the generated output data, subsequent elements of the output data can be generated with higher accuracy because the transformer encoder-decoder can receive more data over time.

[0324] Due to the transformer encoder-decoder architecture, the transformer encoder-decoder can be configured to transform a sequence into another representation of the sequence. An example of transforming a sequence into another representation could be translating a sentence into another language. Various transformer encoder-decoders are available in the art, such as BART, T5, etc.

[0325] In an embodiment, layer normalization 1908, 1912 can be applied before masking multi-head self-attention 1924, multi-head self-attention 1906, and / or feedforward layer 1910 in the transformer decoder, transformer encoder, and / or transformer encoder-decoder. By doing so, computational resources used to apply multi-head self-attention 1906 and / or feedforward layer 1910 to the embedded input data and / or context tensors can be reduced, as fewer entries may be required for the corresponding tensors after normalization.

[0326] In an embodiment, the decoder output 1992 may include a classification neural network, additional feedforward layers, convolutional layers, fully connected layers, etc. For example, the transformer encoder-decoder can be configured to select among multiple options. For this purpose, three different input datasets can be provided to the transformer encoder-decoder, and the context vectors obtained from one or more decoder blocks 1990 can be classified via one or more linear layers. Thus, the architecture can be extended according to the use case to be addressed. [1]

[0327] Figure 20 Examples of training and / or deploying transformer encoders, transformer decoders, and / or transformer encoder-decoders are shown.

[0328] The encoder / decoder / encoder-decoder architecture 2002 can correspond to, as shown in Figures 19A to 19C The transformer decoder, transformer encoder, and / or transformer encoder-decoder described in the context of the transformer decoder, transformer encoder, and / or transformer encoder-decoder.

[0329] The output data generated by the encoder / decoder / encoder-decoder architecture 2002 may include one or more elements, in particular a sequence of elements. Previously generated elements of the output data may be provided as input for generating the next element in the output data sequence.

[0330] exist Figure 20In the example, the input data may include N elements, specifically input lexical units. Input lexical units can be lexical units specifically designed for input into a data-driven model such as a transformer decoder, transformer encoder, or transformer encoder-decoder. The output data to be generated may include M elements. The encoder / decoder / encoder-decoder architecture 2002 can generate one element of the output data in one time step based on the received input data and optionally previously generated output data elements. Therefore, M time steps are required to generate M elements. The time steps include providing inputs 2010, 2012, and 2014 to the encoder / decoder / encoder-decoder architecture 2002 and receiving output data 2004, 2008, and 2006 from the encoder / decoder / encoder-decoder architecture 2002. In the first time step, input 2010 may include N input lexical units. The N input lexical units may, for example, be associated with N words, stems, or word endings. Preferably, the N input lexical units may specify a question. One or more input lexical units may specify the start and / or end of a sequence of lexical units. Input 2010 can be processed by encoder / decoder / encoder-decoder architecture 2002. Based on input 2010, at least a portion of output data 2004 can be generated. At least a portion of the output data may include a first output lexical unit. In the next time step, the generated first output lexical unit may be provided together with input 2012. Specifically, if input 2012 can be received by transformer encoder-decoder, the input lexical unit can be received at encoder input 1988, and the first output lexical unit can be received at decoder input 1994. If input 2012 can be received by transformer encoder, input 2012 can be received at encoder input 1978, and this also applies to transformer decoder and decoder input 1984. Based on input 2012, output data 2008 including a first output lexical unit and a second output lexical unit can be generated. Generating output data 2008 based on input 2012 may mean generating the second lexical unit based on the first lexical unit and N input lexical units, wherein the first lexical unit may have already been generated based on the N input lexical units. This process can be repeated until the last word in the sequence of output data 2006 can be generated. Preferably, the last word can be the end word. The end word can terminate the generation of further output words.

[0331] Similar to the data processing during the deployment of the encoder / decoder / encoder-decoder architecture 2002, the encoder / decoder / encoder-decoder architecture 2002 can be trained. The training dataset can include multiple sequences containing multiple elements. These sequences can be associated with input data and / or output data. Alternatively or additionally, these sequences can be independent of the input data and / or output data. For example, where the input data and output data can refer to chemical compositions represented via text, the training dataset can include sequential text data independent of chemical compositions. In this example, the training dataset can include word sequences derived from a dialogue. In embodiments, the training dataset can at least partially include the input dataset and / or the output dataset.

[0332] Training can be initialized by initializing the encoder / decoder / encoder-decoder architecture 2002. In an embodiment, the parameters associated with the encoder / decoder / encoder-decoder architecture 2002 can be initialized randomly. Alternatively, training can be performed as described in... Figure 18 The CBOW model or skip gram model described in the context is used to obtain the input embeddings of the encoder / decoder / encoder-decoder architecture 2002. The trained embedding layer can be used during training. The parameters associated with the embedding layer can remain unchanged and / or can be updated after a predefined number of training epochs. By doing so, fewer parameters need to be updated, resulting in faster training with less computational resource consumption. Furthermore, the accuracy associated with the embedding layer can remain unchanged and / or can be improved by avoiding error compensation associated with the newly initialized encoder / decoder / encoder-decoder architecture 2002.

[0333] During training of the encoder / decoder / encoder-decoder architecture 2002, at least a portion of the sequence of the training dataset can be provided to the encoder / decoder / encoder-decoder architecture 2002 one after another, and one or more elements can be generated one after another based on the sequence of the training dataset. Elements generated based on the sequence can follow elements of the sequence portions that may be provided to the encoder / decoder / encoder-decoder architecture 2002. The generated one or more elements can be compared with one or more elements as specified by the training dataset that follow at least a portion of the sequence provided to the encoder / decoder / encoder-decoder architecture 2002. Therefore, during training, the encoder / decoder / encoder-decoder architecture 2002 can generate a guess about the next element, and the guess about the next element in the sequence can be compared with the ground truth value of the actual next element specified according to the training dataset. Based on the guess about the next element and the ground truth value, a loss can be determined. The loss can be defined as the similarity between the guess about the next element and the ground truth value. The loss can be determined by forming a vector dot product between the tokens associated with one or more elements and the tokens associated with the ground truth value. When the loss is not zero, it may be necessary to update the parameters associated with the encoder / decoder / encoder-decoder architecture 2002. Preferably, the parameters associated with the encoder / decoder / encoder-decoder architecture 2002 can be independent of the embedding layer. For example, the parameters associated with the encoder / decoder / encoder-decoder architecture 2002 can be the weights of the neurons in the encoder / decoder / encoder-decoder architecture 2002.

[0334] Based on the determined loss, backpropagation can be applied to determine the gradients associated with the parameters of the encoder / decoder / encoder-decoder architecture 2002 to reduce the loss. According to the determined gradients, the parameters associated with the encoder / decoder / encoder-decoder architecture 2002, preferably the weights of the neurons associated with the encoder / decoder / encoder-decoder architecture 2002, can be updated using a gradient descent algorithm.

[0335] The training dataset can be unlabeled. The sequence of elements in the training dataset can inherently include ground truth values ​​for determining the loss based on one or more elements generated during training of the encoder / decoder / encoder-decoder architecture 2002. Therefore, the encoder / decoder / encoder-decoder architecture 2002 can be self-supervised trained. This is advantageous because it saves time and resources spent creating labeled training datasets. Furthermore, it enables the use of large training datasets associated with several megabytes in size. Thus, the data-driven model can be accurate in generating the elements of the sequence. Additionally, the large training dataset enables few-shot or even zero-shot predictions. Therefore, the data-driven model trained as described above is general-purpose and helps save resources required to train and / or host multiple goal-oriented models (such as convolutional neural networks). This training can be referred to as pre-training. The data-driven model can be configured to perform few-shot or even zero-shot predictions for multiple use cases after pre-training. The performance of the data-driven model can be further improved through additional training, known as fine-tuning.

[0336] Figure 21 An example of input embedding is illustrated. Where the sequence of elements associated with the input data (preferably included in the input data) can be of one type, it can be applied as in... Figures 19A to 2 The input embeddings described in the C context are 1902, 1920, 1952, and 1966. For example, the type of input data can be text, where elements can be associated with at least a portion of a word, punctuation characters, a start term indicating the beginning of one or more sequences associated with the input data, and / or an end term. In another example, the input data can be at least partially numeric. Therefore, the input data can include multiple numbers. Numerical input data can be, for example, tabular data. Tabular data can specify one or more rows and / or one or more columns. Therefore, tabular data can include one or more cells, where these cells can be associated with one or more numeric values.

[0337] Numeric input data may require different embeddings than text input data. Input embeddings for numeric input data can include word embeddings, positional embeddings, column embeddings, row embeddings, or combinations thereof.

[0338] Applying lexical embeddings to one or more elements (specifically, lexical units associated with the input data) can produce machine-processable representations associated with those elements (specifically, lexical units). Applying lexical embeddings to one or more elements can instruct those elements through an embedding layer, for example, as in... Figure 18The context is described. Therefore, lexical embeddings can specify one or more elements, specifically lexical terms, in a machine-processable representation. For example, lexical embeddings can transform numerical values ​​into vectors. This is advantageous because such a representation can be enriched with further information, such as the position of the lexical term within the sequence and / or within a table associated with the lexical sequence. Positional embeddings can be similar to... Figure 18 , Figures 19A to 2 Position embeddings are described in the context of C. Column embeddings can be applied when the input data can be tabular data. Applying column embeddings to one or more elements (particularly lexical terms associated with the input data) can produce a machine-processable representation specifying the position of one or more elements within table 2102 (preferably within the columns of table 2102). Applying column embeddings can refer to adding column factors to the input data embedded via lexical embeddings, particularly embedded input data. Column factors can be the same for elements associated with the same column, and / or may differ between two or more elements associated with different columns. Similarly, row embeddings can be applied when the input data can be tabular data. Applying row embeddings to one or more elements (particularly lexical terms associated with the input data) can produce a machine-processable representation specifying the position of one or more elements within table 2102 (preferably within the rows of table 2102). Applying row embeddings can refer to adding column factors to the input data embedded via lexical embeddings, particularly embedded input data. Row factors can be the same for elements associated with the same row, and / or may differ between two or more elements associated with different rows.

[0339] In this embodiment, the input data may be at least partially numerical and at least partially text. Therefore, the input data may include two or more types of data. The data type may refer to a modality. Therefore, different embeddings can be applied to the input data. For the text-containing portions of the input data, embeddings can be applied... Figure 18 , Figures 19A to 2Input embeddings are mentioned in C. For the portion of input data that serves as numeric lexical embeddings, positional embeddings, column embeddings, and row embeddings can be applied. Furthermore, segmented embeddings can be applied to the input data independently of its type. Segmented embeddings can specify the type of input data that can be associated with one or more elements. For example, if the input data includes both text and numbers, the input data can include both types of input data. Applying segmented embeddings to input data can refer to adding segmentation factors to the input data, preferably embedding the input data and / or the input data after which lexical embeddings have been applied. Segmentation factors can specify the type of data associated with one or more elements. Segmentation factors can be the same for one or more elements associated with the same type of input data, and / or may differ between two or more elements associated with different types of input data.

[0340] Applying lexical embedding, positional embedding, segmented embedding, column embedding, row embedding, or combinations thereof can generate embedded input data and / or can be the output of any of encoder inputs 1978, 1984, 1988 or decoder inputs 1984, 1994. Data obtained by applying lexical embedding, positional embedding, segmented embedding, column embedding, row embedding, or combinations thereof can be processed by encoder blocks 1974, 1986, decoder blocks 1980, 1990, encoder output 1976, and decoder outputs 1992, 1982.

[0341] Figure 22 An example of input embedding is shown.

[0342] Data-driven models, especially in Figures 19A to 19C The input data for the encoder input and / or decoder input, as described in the context, may include image data. The data-driven model can be parameterized to receive image data. To process image data as input data, the data-driven model may include, as in... Figures 19A to 19C The context describes one or more encoder blocks and / or one or more decoder blocks and / or one or more encoder outputs and / or one or more decoder outputs. Figure 22 Examples of encoder inputs and / or decoder inputs can be shown. When processing image data, the encoder inputs and / or decoder inputs of the data-driven model can be as follows: Figure 22The encoder input and / or decoder input may include one or more linear projection layers 2214 for linearly projecting one or more images, preferably one or more partial images, more preferably sequences of two or more partial images. The one or more linear projection layers 2214 may be adapted to change the size of one or more received images, preferably one or more partial images, and preferably passing one or more images, preferably partial images through the one or more linear projection layers 2214 may enable the application of image embedding, preferably partial image embedding, to one or more images and / or partial images.

[0343] Furthermore, when a sequence of two or more images and / or partial images can be received, positional embedding can be applied to the sequence, preferably by passing the sequence of one or more images and / or partial images through one or more linear projection layers 2214. Applying positional embedding can refer to adding positional factors. The positional factors can vary depending on the position of the images and / or partial images within the sequence. In particular, the positional factor added to the first element of the sequence can be different from the positional factor added to the second element of the sequence. The first element of the sequence can be a first image and / or a first partial image. The second element of the sequence can be a second image and / or a second partial image.

[0344] A representation of one or more images, preferably one or more portions of an image, can be obtained based on the following equation:

[0345]

[0346] in, It is an image category embedding 2228. It is the nth image in the sequence, specifically a portion of the images. It is a representation of one or more images, preferably one or more partial images, where (H, W) is the resolution of the image, particularly the image on which the partial image is based, C is the number of channels associated with the one or more images, particularly one or more partial images, and D is the dimension of the representation of the one or more images, preferably one or more partial images. Applying partial image embedding can refer to forming With the above equation The product of the above equations. Applying positional embedding can refer to adding factors according to the above equation. .

[0347] By doing so, text-based data, numerical data, tabular data, image data, and so on can be processed by a data-driven model.

[0348] This disclosure has also been described in conjunction with various preferred embodiments and examples. However, by studying the accompanying drawings, this disclosure, and the claims, those skilled in the art, as well as those practicing the claimed subject matter, will understand and implement other variations. It is particularly noteworthy that any steps presented can be performed in any order; that is, this disclosure is not limited to a specific order of these steps. Furthermore, it is not required that different steps be performed at a specific location or node in a distributed system; that is, each step can be performed on different nodes using different devices / data processing.

[0349] As used herein, "determine" also includes "initiating or causing determination," "generate" also includes "initiating and / or causing generation," and "provide" also includes "initiating or causing determination, generation, selection, sending, and / or receiving." "Initiating or causing an action" includes any processing signal that triggers a computing node or device to perform a corresponding action.

[0350] In the claims and specification, the word "comprising" or "including" or similar wording does not exclude other elements or steps and should not be construed as limiting oneself to the listed elements or steps. The indefinite article "a" or "an" does not exclude multiple. A single element or other unit may perform the function of several entities or items recited in the claims. The fact that certain measures are recited only in mutually different dependent claims does not indicate that a combination of these measures cannot be used in advantageous implementations or that additional elements may be included.

[0351] Within the scope of this disclosure, provision may include any interface configured to provide data. This may include application programming interfaces, human-machine interfaces (such as displays), and / or software module interfaces. Provision may include transmitting or submitting data to the interface, particularly displaying data to a user or having data used by a receiving entity.

[0352] Any disclosures and embodiments described herein relate to the methods, systems, devices, and computer program elements listed above, and vice versa. Advantageously, the benefits provided by any embodiment and example also apply to all other embodiments and examples, and vice versa.

[0353] This disclosure has also been described in conjunction with various preferred embodiments and examples. However, by studying the accompanying drawings, this disclosure, and the claims, those skilled in the art, as well as those practicing the claimed invention, will understand and implement other variations. It is particularly noteworthy that any of the proposed steps can be performed in any order; that is, the invention is not limited to a specific order of these steps. Furthermore, it is not required that different steps be performed at a specific location or node in a distributed system; that is, each step can be performed on different nodes using different devices / data processing.

[0354] As used herein, "determine" also includes "initiating or causing determination," "generate" also includes "initiating and / or causing generation," and "provide" also includes "initiating or causing determination, generation, selection, sending, and / or receiving." "Initiating or causing an action" includes any processing signal that triggers a computing node or device to perform a corresponding action.

[0355] In the claims and description, the word "comprising" does not exclude other elements or steps. The indefinite articles "a" or "an" and the definite article "the" do not exclude multiple. In particular, the indefinite article "a" or "an" can be replaced by one or more, and the definite article "the" can also be replaced by one or more. A single element or other unit can perform the function of several entities or items recited in the claims. The fact that certain measures are recited only in mutually different dependent claims does not indicate that a combination of these measures cannot be used in advantageous implementations.

[0356] Any disclosures and embodiments described herein relate to the methods, systems, devices, and computer program elements listed above, and vice versa. Advantageously, the benefits provided by any embodiment and example also apply to all other embodiments and examples, and vice versa.

Claims

1. A method, particularly a computer-implemented method, for generating chemical product data characterizing chemical products having one or more target properties, comprising: A request is provided for providing a target chemical product, wherein the request includes instructions regarding the target chemical product. Provide functional specification data related to one or more functions of one or more operating engines used to provide data on the chemical product. Provide one or more input data structures that are suitable for the input data provided to the one or more operating engines. Task instructions related to the request, the one or more input data structures, and the functional specification data are provided to one or more data-driven models, wherein the one or more data-driven models are configured to generate operational input data for one or more selected operational engines in response to the provided instructions regarding the target chemical product. The operational input data includes structured data used to trigger at least one selected representation operational engine to provide a digital representation of the chemical structure of the target chemical product and, based on the digital representation of the chemical structure of the target chemical product, provide at least a portion of the chemical product data. The operation input data is provided to the at least one selected operation engine to provide at least a portion of the chemical product data. Provide data on the generated chemical products for the production and / or processing of target chemical products having one or more of the target properties.

2. The method according to claim 1, wherein, The task instructions provided include: The representation task instructions, including unstructured data, associated with the request, the one or more input data structures, and the functional specification are provided to one or more data-driven models, wherein the one or more data-driven models are configured to generate representation operation input data for one or more selected representation operation engines in response to the provided instructions regarding the target chemical product, and wherein the representation operation input data includes structured data used to trigger the selected representation operation engine to provide a digital representation of the chemical structure of the target chemical product. The representation operation input data is provided to the at least one selected representation operation engine to provide a digital representation of the chemical structure of the target chemical product. The data generation task instructions related to the digital representation of the chemical structure of the target chemical product, the request, and the functional specification data are provided to the one or more data-driven models. These data-driven models can be further configured to generate data generation operation input data for one or more selected data generation operation engines in response to the provided instructions regarding the target chemical product. The data generation operation input data may include structured data used to trigger the selected data generation operation engine to provide at least a portion of the chemical product data. The input data for the data generation operation is provided to the selected data generation operation engine to provide at least a portion or a combination of the chemical product data.

3. The method according to claim 2, wherein, The representation operation engine is a database configured to provide a digital representation of the target chemical product in response to a structured query relating to an indication of the target chemical product, wherein the representation operation input data includes the structured query.

4. The method according to claim 2 or 3, wherein, The data generation operation engine is a database configured to provide at least a portion of the chemical product data in response to a structured query relating to an indication of the target chemical product, wherein the data generation operation input data includes the structured query, and / or wherein the data generation operation engine is configured to determine a model, particularly a data-driven model and / or a physical model of the chemical product data according to a provided data generation task instruction. The physical model may include one or more equations to generate the chemical product data according to the data generation task instruction. The physical model may be associated with one or more equations relating to a functional dependency between the chemical product data and / or the data generation task instruction. This functional dependency may be based on one or more mathematical equations that define a functional relationship between one or more measures associated with the chemical product data and one or more measures associated with the data generation task instruction.

5. The method according to any one of claims 1 to 4, wherein, The one or more data-driven models include a representation data-driven model configured to generate representation operation input data for one or more selected representation operation engines in response to a provided instruction regarding the target chemical product, and wherein the one or more data-driven models further include a data generation data-driven model configured to generate data generation operation input data for one or more selected data generation operation engines in response to a provided instruction regarding the target chemical product, and wherein a representation task instruction is provided to the representation data-driven model, and wherein the data generation task instruction is provided to the data generation data-driven model.

6. The method according to any one of claims 1 to 5, wherein, The one or more data-driven models include pre-trained data-driven models, wherein the pre-trained data-driven models are configured to perform multiple different tasks according to multiple different task instructions. In embodiments, the representation data-driven model and / or the data generation data-driven model may be pre-trained data-driven models.

7. The method according to any one of claims 1 to 6, wherein, The one or more data-driven models include a fine-tuned data-driven model, wherein the fine-tuned data-driven model is obtained by further training a pre-trained data-driven model based on training data, which includes task instructions and corresponding operation input data, wherein the pre-trained data-driven model is configured to perform multiple different tasks according to multiple different task instructions.

8. The method according to any one of claims 1 to 7, wherein, The one or more data-driven models are further configured to map the numerical representation of the task instruction to a context task instruction, and to map the context task instruction to a numerical representation of the operation input data, wherein the context task instruction is obtained by processing the numerical representation of the task instruction through one or more matrix operations associated with the one or more data-driven models.

9. The method according to any one of claims 1 to 8, wherein, Providing the task instructions, and in particular indicating task instructions and / or data generation task instructions to generate the input data for the operation, includes: The selection task, including unstructured data, related to the request and the functional specification data is provided to the selection model, which is configured to generate model output data related to the selected operating engine and the one or more target features. The model output data includes unstructured data, and... The generated model output data is provided to the structural model, which is configured to generate operational input data based on the model output data.

10. The method according to any one of claims 1 to 9, further comprising providing verification task instructions related to the operation input data, the at least one selected operation engine, and the one or more input data structures to the one or more data-driven models to verify the data structures related to the operation input data, wherein, The one or more data-driven models are further configured to classify whether the data structure associated with the operation input data corresponds to the input data structure associated with the at least one selected operation engine.

11. The method according to any one of claims 1 to 10, wherein, Providing the task instruction includes mapping the task instruction to a numerical representation of the task instruction, wherein the one or more data-driven models are configured to map the numerical representation of the task instruction to a numerical representation of the operational input data.

12. An apparatus comprising: processor; as well as A memory storing instructions that, when executed by the processor, configure the device to perform the steps of any one of the methods as claimed in any one of claims 1 to 11.

13. Use of chemical product data obtained by any one of claims 1 to 11 for the production and / or processing of chemical products.

14. The use of the task instruction according to any one of claims 1 to 11 for processing a request for providing chemical product data to produce and / or process the target chemical product according to any one of claims 1 to 11.

15. The use of one or more data-driven models according to any one of claims 1 to 11 for providing chemical product data for the production and / or processing of target chemical products.

Images