Apparatus, data structure and computer-implemented method for determining an orchestration of tasks

Abstract

A device, a data structure and a computer-implemented method for determining an orchestration of tasks, the method comprising receiving (202) names of tasks and sources for executing the tasks, in particular at an interface, and determining (204) the orchestration of the tasks with a model for generative natural language processing depending on the names of the tasks and sources and a description of tasks and sources in natural language.

Description

background

[0001] The invention relates to a device, a data structure and a computer-implemented method for determining an orchestration of tasks.

[0002] Generative artificial intelligence systems, such as large language models, are capable of orchestrating and executing artificial intelligence tasks. Disclosure of the invention

[0003] A computer-implemented method for determining an orchestration of tasks comprises receiving names of tasks and sources for executing the tasks, in particular at an interface, and determining the orchestration of the tasks with a generative natural language processing model depending on the names of the tasks and sources and a natural language description of the tasks and sources. The natural language description of the tasks and sources conveys domain knowledge. The method, which uses the generative natural language model with this domain knowledge, mitigates hallucination and lack of grounding.The method for determining the orchestration with the generative natural language model based on the description of tasks and sources in natural language can be part of a hybrid generative artificial intelligence process that includes the method for determining an orchestration of tasks and a method for executing the tasks based on the sources according to the orchestration. In this context, the method for determining the orchestration significantly improves the efficiency of innovation pipelines.

[0004] For example, at least one of the tasks includes collecting simulation data, collecting test run data, collecting regularities, collecting standard data, collecting domain knowledge, in particular human feedback, identifying a simulation tool for performing a simulation, and identifying an optimization application interface for hyperparameter optimization and / or performing a literature review.

[0005] For example, at least one of the sources includes simulation data, test run data, regularities, standard data, domain knowledge, in particular human feedback, a simulation tool for performing a simulation, an optimization application interface for hyperparameter optimization and / or literature, in particular private or publicly available scientific publications or know-how.

[0006] The method may include receiving a mapping of at least one of the names of sources to at least one name of tasks and determining orchestration with the generative natural language processing model depending on the mapping.

[0007] The process can interact with a user or with a machine in an innovation process.

[0008] The method may, for example, include prompting the user or machine to enter the association.

[0009] The method may, for example, include prompting the user or machine to enter the names of tasks and sources.

[0010] The method according to one of the preceding claims, characterized in that the orchestration of tasks comprises code, in particular machine-interpretable or machine-executable code, which, when the code is executed by the machine, orchestrates the tasks, wherein determining the orchestration comprises determining the code. The orchestration of tasks is a recurring part of daily research and development work. The method uses domain know-how to support this work by automating the repetitive part of the orchestration and generating code that can be used to trigger the execution of the tasks according to the orchestration.

[0011] The method may include executing the code. This means that the method orchestrates the execution of the tasks. This enables, for example, the generation of a code prototype with the tasks.

[0012] A device for determining an orchestration of tasks comprises at least one processor, at least one memory, and at least one interface, wherein the at least one memory is configured to store instructions that, when executed by the at least one processor, cause the device to perform the method. The device for determining the orchestration can be part of a hybrid generative artificial intelligence system that includes the device for determining the orchestration and is configured to execute the tasks according to the orchestration.

[0013] A computer program may comprise instructions, the instructions being executable by a computer, and the instructions, when executed by the computer, causing the computer to perform the method.

[0014] A data structure, in particular a computer-implemented data structure, for determining an orchestration of tasks can comprise at least one data field for names of tasks and sources for executing the tasks, at least one data field for the orchestration of the tasks.

[0015] The data structure may comprise at least one data field for mapping at least one of the names of sources to at least one name of tasks.

[0016] Further embodiments are apparent from the following description and the drawing. In the drawing: Fig. 1 schematically shows a device for determining an orchestration of tasks, Fig. 2 shows a flowchart with steps of a method for determining the orchestration of tasks, Fig. Figure 3 schematically shows a data structure for determining the orchestration of tasks.

[0017] Fig. 1 schematically shows an apparatus 100 for determining an orchestration of tasks.

[0018] The device 100 includes at least one processor 102, at least one memory 104, and at least one interface 106. The at least one memory 104 may include a non-volatile memory. The at least one memory 104 may include a volatile memory.

[0019] The at least one memory 104 is configured to store instructions that, when executed by the at least one processor 102, cause the device 100 to perform a method for orchestrating tasks.

[0020] According to one example, the at least one memory 104 is configured to store a generative natural language processing model. According to one example, the generative natural language processing model has access to a natural language description of tasks and sources. The at least one memory 104 may be configured to store the natural language description of tasks and sources. In this context, access may mean that the generative natural language processing model has access to text comprising the natural language description of tasks and sources. In this context, access may mean that the generative natural language processing model has access to embeddings of text comprising the natural language description of tasks and sources.

[0021] The generative natural language processing model is trained to determine the orchestration of tasks based on task and source names and on the natural language description of tasks and sources. This means that an input to the generative natural language processing model depends on the names of the tasks and the names of the sources.

[0022] In this context, determining the orchestration of tasks depending on task names and source names may mean that the natural language processing model is configured to process text that includes the names. In this context, determining the orchestration of tasks depending on task names and source names may mean that the natural language processing model is configured to process embeddings of text that includes the names.

[0023] The interface 106 is configured to receive the names of tasks and sources for executing the tasks. The interface 106 is configured, for example, to prompt a user or machine to enter the tasks and sources. The interface 106 is configured, for example, to prompt the user or machine to enter a mapping of sources and tasks. To prompt the user, the interface 106 comprises, for example, a graphical user interface. To prompt the machine, the interface 106 comprises, for example, a machine-to-machine interface.

[0024] The device 100 is configured to determine the orchestration of the tasks with the generative natural language processing model depending on the names of the tasks and sources and the description of the tasks and sources in natural language. This means that the orchestration depends on an output of the generative natural language processing model.

[0025] According to one example, the orchestration of tasks comprises code, in particular code that is interpretable or executable by a machine. According to one example, the orchestration comprises code that orchestrates the tasks when the code is executed by the machine.

[0026] The device 100 may be configured to determine the code using the model for generative natural language processing.

[0027] In this context, determining the orchestration or code may mean that the natural language processing model is configured to output text comprising the orchestration or code. In this context, determining the orchestration or code may mean that the natural language processing model is configured to output embeddings of text comprising the orchestration or code.

[0028] The apparatus 100 may be configured to execute the code, e.g., to orchestrate the execution of the tasks depending on the sources associated with the tasks and in accordance with the orchestration.

[0029] In this context, orchestration refers, for example, to an order in which tasks are executed. Orchestration refers, for example, to a dependency between tasks when executing tasks. Orchestration refers, for example, to a constraint on the execution of tasks. The constraint can depend on the result of a task's execution.

[0030] Orchestration can include the order in which tasks are executed. Orchestration can include the assignment of sources to tasks for executing the tasks to the sources assigned to the tasks. Orchestration can include dependencies. Orchestration can include constraints.

[0031] Example tasks are listed below: Task 1: Collecting simulation data, Task 2: Collecting test run data, Task 3: Collecting regularities, Task 4: Collecting standard data, Task 5: Collecting domain knowledge, especially human feedback, Task 6: Identify a simulation tool to perform a simulation, Task 8: Identifying an optimization application interface for hyperparameter optimization, Task 9: Conduct a literature review.

[0032] Example sources are listed below: Source 1: Simulation data, Source 2: Test run data, Source 3: Laws, Source 4: Standard data, Source 5: Domain knowledge, especially human feedback, Source 6: a simulation tool for performing a simulation, Source 7: an optimization application interface for hyperparameter optimization, Source 8: Literature, in particular private or publicly accessible scientific publications or know-how.

[0033] The sources can be deeply rooted knowledge, e.g., static, pre-calculated data.

[0034] The sources can be stored in the at least one memory 104. The sources can be accessed via the interface 106. The sources can be stored in databases, text documents, specifications, invention reports, and production documentation.

[0035] The sources can be collected on demand. The sources can be collected by exploration. The sources can be collected by computing data on demand or by exploring data. The data characterizes a digital twin or digital twins of a factory or product. The data can characterize a technical system or a simulation of a technical system. The data can characterize an artificial intelligence service. The data can characterize one or more knowledge graphs. The data can be received from a digital twin or digital twins of a factory or product. The data can be received from a technical system or a simulation of a technical system. The data can be received from an artificial intelligence service. The data can be received from a knowledge graph or knowledge graphs.

[0036] The technical system may be a robot, a vehicle, a household appliance, a power tool, a personal assistance system or a sensor or control component for a robot, a vehicle, a household appliance, a power tool, a personal assistance system or an access control system.

[0037] The simulation data or test run data may include audio data, digital images, e.g., video, radar, LiDAR, ultrasound, motion, thermal images

[0038] The control component may be an electronic stability program or an autonomous driving system. A digital image may include an infrastructure object, e.g., a traffic sign, a road surface, and / or a road user, e.g., a pedestrian or a vehicle.

[0039] An example orchestration might include: - Perform task 5 subject to task 6, within the constraints of task 3, subject to task 1, and subject to task 4, with the constraint that an improvement is observed when comparing the result of task 5 with the result of task 2. - Complete task 7

[0040] Fig. Figure 2 shows a flowchart with steps of the method for determining the orchestration of tasks. A computer program may include the instructions, wherein the instructions are executable by a computer. When executed by the computer, the instructions cause the computer to perform the method.

[0041] The method includes a step 202.

[0042] Step 202 includes receiving names of tasks and sources for executing the tasks.

[0043] For example, the names of tasks and sources are received via interface 106.

[0044] The method may include prompting a user or machine to enter the names of tasks and sources. The user or machine may provide the tasks according to a task schedule, i.e., a plan specifying which tasks are required for a project. The user or machine may provide the sources according to a source selection, i.e., a selection of sources specifying which sources are required for a project.

[0045] The method may include receiving a mapping of at least one of the names of sources to at least one name of tasks.

[0046] The method may include prompting the user or machine to enter the mapping. The user or machine may provide the mapping according to the user's or machine's domain expertise. The mapping may associate a resource with the task required to perform a task.

[0047] The method includes a step 204.

[0048] In step 204, the orchestration is determined. The orchestration is determined using the generative natural language processing model depending on the names of tasks and sources and depending on the natural language description of tasks and sources.

[0049] According to one example, the orchestration is additionally determined depending on the assignment.

[0050] The code is determined according to an example.

[0051] Optionally, the method includes a step 206.

[0052] Step 206 involves executing the code.

[0053] Executing the code may involve calling the tasks in accordance with the orchestration.

[0054] Executing the code may include issuing, e.g., via interface 106, instructions to invoke the tasks in accordance with the orchestration.

[0055] Data accessed by the device 100 or method may be provided in a data structure for orchestrating tasks.

[0056] Fig. 3 shows an example of a data structure 300 for determining an orchestration of tasks.

[0057] For example, data structure 300 is a computer-implemented data structure.

[0058] The data structure 300 includes at least one data field 302 for names of tasks and sources for executing the tasks.

[0059] The data structure 300 includes at least one data field 302 for the orchestration of the tasks.

[0060] The data structure 300 may include at least one data field 302 for mapping at least one of the names of sources to at least one name of tasks.

Claims

[1] Computer-implemented method for determining an orchestration of tasks, characterized by that the method comprises receiving (202) names of tasks and sources for executing the tasks, in particular at an interface, and determining (204) the orchestration of the tasks with a model for generative natural language processing depending on the names of the tasks and sources and a description of tasks and sources in natural language. [2] Method according to one of the preceding claims, characterized bythat at least one of the tasks comprises collecting simulation data, collecting test run data, collecting regularities, collecting standard data, collecting domain knowledge, in particular human feedback, identifying a simulation tool for performing a simulation, identifying an optimization application interface for hyperparameter optimization, and / or performing a literature review. [3] Method according to one of the preceding claims, characterized by that at least one of the sources comprises simulation data, test run data, regularities, standard data, domain knowledge, in particular human feedback, a simulation tool for performing a simulation, an optimization application interface for hyperparameter optimization and / or literature, in particular private or publicly available scientific publications or know-how. [4] Method according to claim 1, characterized bythat the method comprises receiving (202) an assignment of at least one of the names of sources to at least one name of tasks and determining (204) the orchestration with the generative natural language processing model depending on the assignment. [5] Method according to claim 3, characterized by that the method comprises requesting (202) a user or a machine to enter the association. [6] Method according to one of the preceding claims, characterized by that the method comprises prompting (202) a user or machine to enter names of tasks and sources. [7] Method according to one of the preceding claims, characterized bythat the orchestration of tasks comprises code, in particular machine-interpretable or machine-executable code, which, when the code is executed by the machine, orchestrates the tasks, wherein determining (204) the orchestration comprises determining the code. [8] Method according to claim 7, characterized by that the method comprises executing (206) the code. [9] Device (100) for determining an orchestration of tasks, characterized by in that the device (100) comprises at least one processor (102), at least one memory (104) and at least one interface (106), wherein the at least one memory (104) is designed to store instructions which, when executed by the at least one processor (102), cause the device (100) to carry out the method according to one of claims 1 to 8. [10] Computer program, characterized byin that the computer program comprises instructions, the instructions being executable by a computer, and the instructions, when executed by the computer, causing the computer to carry out the method according to any one of claims 1 to 8. [11] Data structure (300), in particular computer-implemented data structure, for determining an orchestration of tasks, characterized by that the data structure (300) comprises at least one data field (302) for names of tasks and sources for executing the tasks, at least one data field for the orchestration of the tasks. [12] Data structure according to claim 11, characterized by that the data structure (300) comprises at least one data field (302) for assigning at least one of the names of sources to at least one name of tasks.

Images