Knowledge graph construction via generative artificial intelligence

Abstract

Systems or techniques are provided that facilitate knowledge graph construction via generative artificial intelligence. In various embodiments, a system may access a plurality of electronic documents (e.g., 106) associated with the design or manufacture of a medical imaging scanner (e.g., 104). In various aspects, the system may construct a knowledge graph (e.g., 206) representing the plurality of electronic documents by iteratively executing a generative text-to-text neural network (e.g., 202) on a design discovery tree (e.g., 204) associated with the medical imaging scanner. In various examples, the system may access a natural language query (e.g., 1602) regarding the medical imaging scanner, and may convert the natural language query to a structured query (e.g., 1606) via execution of another neural network (e.g., 1604). In various cases, the system may execute the structured query on the knowledge graph, thereby generating an electronic answer (e.g., 1608) to the natural language query.

Description

Technical Field

[0001] The present disclosure generally relates to knowledge graphs, and more particularly, to knowledge graph construction via generative artificial intelligence. Background Art

[0002] Medical imaging scanners can be deployed on-site. During deployment, a medical imaging scanner may malfunction. The user or operator of the medical imaging scanner can determine the cause of such a malfunction or the way to troubleshoot such a malfunction by consulting a knowledge graph corresponding to the medical imaging scanner. Unfortunately, the existing technologies for constructing such knowledge graphs are domain-specific and thus cannot be easily implemented across different domains.

[0003] Therefore, a system or technology that can solve one or more of these technical problems may be desirable. Summary of the Invention

[0004] The following gives a summary of the invention to provide a basic understanding of one or more embodiments. This summary of the invention is not intended to identify key or important elements, nor is it intended to delineate any scope of a particular embodiment or any scope of the claims. Its sole purpose is to present concepts in a simplified form as a prelude to the more detailed description that is presented later. In one or more embodiments described herein, devices, systems, computer-implemented methods, apparatuses, or computer program products that facilitate knowledge graph construction via generative artificial intelligence are described.

[0005] According to one or more embodiments, a system is provided. The system may include a non-transitory computer-readable memory that may store computer-executable components. The system may also include a processor that may be operatively coupled to the non-transitory computer-readable memory and may execute the computer-executable components stored in the non-transitory computer-readable memory. In various embodiments, the computer-executable components may include an access component that may access a plurality of electronic documents associated with the design or manufacture of a medical imaging scanner. In various aspects, the computer-executable components may include a graph component that may construct a knowledge graph representing the plurality of electronic documents by iteratively executing a generative text-to-text neural network on a design discovery tree associated with the medical imaging scanner. In various instances, the access component may access a natural language query regarding the medical imaging scanner, and the computer-executable components may include a query component that may convert the natural language query into a structured query via the execution of another neural network and may execute the structured query on the knowledge graph, thereby generating an electronic answer to the natural language query.

[0006] According to one or more embodiments, a computer-implemented method is provided. In various embodiments, the computer-implemented method may include: accessing, by a device operatively coupled to a processor, a plurality of electronic documents associated with the design or manufacture of a medical imaging scanner. In various aspects, the computer-implemented method may include: constructing, by the device, a knowledge graph representing the plurality of electronic documents by iteratively performing a generative text-to-text neural network on a design discovery tree associated with the medical imaging scanner. In various instances, the computer-implemented method may include: accessing, by the device, a natural language query regarding the medical imaging scanner. In various cases, the computer-implemented method may include: converting, by the device, the natural language query into a structured query via the execution of another neural network. In various aspects, the computer-implemented method may include: performing, by the device, the structured query on the knowledge graph, thereby generating an electronic answer to the natural language query.

[0007] According to one or more embodiments, a computer program product for facilitating knowledge graph construction via generative artificial intelligence is provided. In various embodiments, the computer program product may include a non-transitory computer-readable memory having program instructions embodied thereon. In various aspects, the program instructions may be executable by a processor to cause the processor to access one or more electronic documents associated with the design or manufacture of a machine. In various instances, the program instructions may be further executable by the processor to cause the processor to construct a knowledge graph representing the plurality of electronic documents by iteratively performing a generative text-to-text neural network on a design discovery tree associated with the machine. In various cases, the program instructions are executable by the processor to further cause the processor to access a natural language query regarding the machine and convert the natural language query into a structured query via the execution of another neural network. In various aspects, the program instructions are executable by the processor to further cause the processor to perform the structured query on the knowledge graph, thereby generating an electronic answer to the natural language query. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 A block diagram illustrating an example non-limiting system for facilitating knowledge graph construction via generative artificial intelligence in accordance with one or more embodiments described herein.

[0009] Figure 2 A block diagram illustrating an example non-limiting system for facilitating knowledge graph construction via generative artificial intelligence in accordance with one or more embodiments described herein, the system including a deep learning neural network, a design discovery tree, and a knowledge graph.

[0010] Figure 3 A block diagram illustrating an example non-limiting design discovery tree in accordance with one or more embodiments described herein.

[0011] Figures 4 to 12 An example non-limiting block diagram illustrating how a knowledge graph may be constructed via generative artificial intelligence according to one or more embodiments described herein is illustrated.

[0012] Figures 13 to 15 A flowchart illustrating an example non-limiting computer-implemented method for facilitating knowledge graph construction via generative artificial intelligence according to one or more embodiments described herein.

[0013] Figure 16 A block diagram of an example non-limiting system that facilitates knowledge graph construction via generative artificial intelligence according to one or more embodiments described herein is illustrated, the system comprising a natural language query, another deep learning neural network, a structured query, and an answer.

[0014] Figure 17 An example non-limiting block diagram illustrating how a natural language query may be converted to a structured query and then executed on a knowledge graph in accordance with one or more embodiments described herein is illustrated.

[0015] Figure 18 A block diagram of an example non-limiting system that facilitates knowledge graph construction via generative artificial intelligence according to one or more embodiments described herein is illustrated, the system including a training component.

[0016] Figure 19 An example non-limiting block diagram showing how a deep learning neural network may be trained is illustrated in accordance with one or more embodiments described herein.

[0017] Figure 20 A flowchart illustrating an example non-limiting computer-implemented method for facilitating knowledge graph construction via generative artificial intelligence according to one or more embodiments described herein.

[0018] Figure 21 A block diagram illustrating an example non-limiting operating environment in which one or more embodiments described herein may be facilitated.

[0019] Figure 22 An example networking environment is illustrated that is operable to perform various implementations described herein. DETAILED DESCRIPTION

[0020] The following detailed description is merely illustrative and is not intended to limit the embodiments or the application / use of the embodiments. In addition, it is not intended to be bound by any express or implied information set forth in the aforementioned "background technology" or "invention content" section or "detailed description" section.

[0021] Reference is now made to the drawings, in which like reference numerals are used throughout to designate like elements. In the following description, numerous specific details are set forth for purposes of explanation in order to provide a more thorough understanding of one or more embodiments. It will be evident, however, that in various instances, one or more embodiments may be practiced without these specific details.

[0022] Medical imaging scanners (e.g., computed tomography (CT) scanners, magnetic resonance imaging (MRI) scanners, X-ray scanners, ultrasound scanners, positron emission tomography (PET) scanners, nuclear medicine (NM) scanners) may be deployed on-site. Accordingly, a medical imaging scanner may capture or generate real-world medical scan images (e.g., CT scan images, MRI scan images, X-ray scan images, ultrasound scan images, PET scan images, NM scan images) of a real-world medical patient (e.g., a human, an animal, or other object).

[0023] During deployment, a medical imaging scanner may malfunction. For example, a medical imaging scanner may experience a hardware failure or a software failure that impedes or prevents the medical imaging scanner from correctly capturing or generating a medical scan image (e.g., the medical scan image may be corrupted or affected by artifacts such as blur or distortion).

[0024] A user or operator of a medical imaging scanner may determine the cause of such a malfunction or a way to troubleshoot such a malfunction by consulting a knowledge graph corresponding to the medical imaging scanner. A knowledge graph may be a graph data structure having nodes and edges, where the nodes represent respective technical features of the medical imaging scanner and where the edges represent respective relationships between the nodes. Structured queries regarding the medical imaging scanner may be answered by performing on the knowledge graph. Such a performance may be carried out in any suitable manner, such as via SQL (an acronym representing "Structured Query Language"), SPARQL (a recursive acronym representing "SPARQL Protocol and RDF Query Language"), GraphQL (a partial acronym representing "Graphical Query Language"), or the Cypher query language.

[0025] Various prior arts construct such knowledge graphs via a text-to-graph parser (also known as an Abstract Meaning Representation (AMR) parser). In particular, a text-to-graph parser is a machine learning model (e.g., a deep learning neural network) that receives a text input and produces as output a graph data structure that semantically represents the text input. Thus, when implementing the prior art, electronic documents describing technical information about a medical imaging scanner are collected, and the text-to-graph parser is performed on such electronic documents, thereby producing a knowledge graph whose nodes and edges represent the technical information of the medical imaging scanner.

[0026] Unfortunately, such existing technologies are highly domain-specific and thus cannot be readily or easily implemented across different technical domains. In fact, in order for a text-to-graph parser to accurately generate a knowledge graph based on a collection of electronic documents, it must be extensively trained on any technical domain to which those electronic documents belong. However, after such extensive training (which can be labor-intensive and time-consuming), the text-to-graph parser may not be able to accurately generate a knowledge graph for electronic documents belonging to other technical domains.

[0027] For example, assume that a text-to-graph parser is trained to accurately create a knowledge graph based on electronic documents related to CT scanners. That is, the text-to-graph parser can be considered specific to the CT domain. In this case, the text-to-graph parser may not be able to accurately generate a knowledge graph based on, instead, electronic documents related to MRI scanners or PET scanners. In other words, a text-to-graph parser that has been trained with respect to the CT domain does not function accurately in the MRI domain or the PET domain. In still other words, the text-to-graph parser does not have generality outside of the technical domain on which it has been trained (at least not without extensive retraining).

[0028] Accordingly, a system or technique that can solve one or more of these technical problems may be desirable.

[0029] The various embodiments described herein can address one or more of these technical problems. One or more embodiments described herein can include a system, a computer-implemented method, an apparatus, or a computer program product that can facilitate knowledge graph construction via generative artificial intelligence. In particular, the inventors of the various embodiments described herein have devised various techniques that enable a knowledge graph to be constructed by a generative text-to-text model. A generative text-to-text model can be a machine learning model (e.g., a deep learning neural network) that can receive input text and output synthetic text that is semantically based on such input text. As described herein, when given a collection of electronic documents that convey technical information about a medical imaging scanner, the inventors of the present invention recognized that a knowledge graph representing the collection of electronic documents can be created by iteratively performing a generative text-to-text model on a design discovery tree associated with the medical imaging scanner. More specifically, the knowledge graph can initially be empty, and the design discovery tree can be an ordered hierarchical structure of defined text prompts that have known relationships with each other, where each text prompt can be a semantic question (e.g., an interrogative sentence) that generally or generically asks some corresponding technical details about the medical imaging scanner (e.g., asking about the component subsystems of the medical imaging scanner, asking about the possible error codes of the component subsystems, asking about the possible diagnostic tests for resolving the possible error codes). For any given text prompt in the design discovery tree, the generative text-to-text model can be performed on the given text prompt in a retrieve augmented generation (RAG) manner with reference to the collection of electronic documents. Such performance can produce some synthetic text content that can be considered to answer the given text prompt with any appropriate information conveyed or described by the collection of electronic documents. In various instances, an empty node shell representing the given text prompt can be inserted into the knowledge graph, one or more edges respectively representing any known relationships associated with the given text prompt can be attached to the empty node shell, and the empty node shell can be filled with the synthetic text content. This can be repeated for each text prompt in the design discovery tree, which can cause the knowledge graph to be iteratively expanded such that its edges represent the known relationships indicated by the design discovery tree and such that its nodes represent any substantial technical information from the collection of electronic documents that answers the text prompts of the design discovery tree.

[0030] The various embodiments described herein can be considered superior to the prior art. In fact, the inventors of the present invention recognize that generative text-to-text models (e.g., such as ChatGPT) can exhibit a broader or wider generality than text-to-graph models. In other words, generative text-to-text models can have a higher tendency to accurately or reliably perform text synthesis on the input text belonging to various different technical fields. Therefore, generative text-to-text models can accurately perform across different technical fields with little fine-tuning, while text-to-graph models cannot accurately perform across different technical fields without extensive retraining. In addition, the design discovery tree can be general enough to be applicable across different technical fields. However, even in cases where a domain-specific design discovery tree is required, the creation of such a domain-specific design discovery tree can be much less time-consuming and labor-intensive than extensively retraining a text-to-graph model. Therefore, compared with the prior art, the various embodiments described herein can be considered a more general way of constructing a knowledge graph.

[0031] The various embodiments described herein can be considered to facilitate computerized tools (e.g., any suitable combination of computer-executable hardware or computer-executable software) for knowledge graph construction via generative artificial intelligence. In various aspects, such computerized tools can include an access component, a graph component, or a query component.

[0032] In various embodiments, there can be a medical imaging scanner. In various aspects, the medical imaging scanner can be any suitable type of medical image capture equipment or modality (e.g., CT scanner, MRI scanner, X-ray scanner, ultrasound scanner, or PET scanner, NM scanner, etc.). In various cases, the medical imaging scanner can be deployed in any suitable clinical or operational environment (e.g., in a hospital, in a veterinary clinic, on an emergency vehicle). In various cases, the medical imaging scanner can include any suitable human-machine interface device (e.g., keyboard, keypad, touch screen, voice command system).

[0033] In various embodiments, a medical imaging scanner may be associated with multiple design documents. In various aspects, each of the multiple design documents may be any suitable electronic file (e.g., a service manual or guide, a bill of materials, a manufacturing instruction set, a blueprint or schematic, a failure mode analysis report) that describes or otherwise indicates in a textual manner (or in some cases, graphically or numerically) any suitable information related to the technical design or manufacture of the medical imaging scanner. In various instances, any of the multiple design documents may be written by or have been written by a technician or engineer responsible for designing, developing, prototyping, correcting, or producing the medical imaging scanner (e.g., via any suitable word processing software, computer-aided design software, or quantitative analysis software).

[0034] In any case, it may be desirable to generate a knowledge graph representing the multiple design documents. As described herein, a computerized tool may facilitate this determination of knowledge graph generation.

[0035] In various embodiments, an access component of the computerized tool may electronically access the multiple design documents. For example, the access component may electronically retrieve or obtain the multiple design documents from any suitable centralized or decentralized database, whether remote or local to the access component. In any case, the access component may electronically access the multiple design documents such that other components of the computerized tool may electronically interact with the multiple design documents (e.g., read, write, edit, copy, manipulate).

[0036] In various embodiments, a graph component of the computerized tool may electronically store, maintain, control, or otherwise access a first deep learning neural network. In various aspects, the first deep learning neural network may exhibit any suitable internal architecture. For example, the first deep learning neural network may include any suitable number of any suitable type of layer (e.g., an input layer, one or more hidden layers, an output layer, any of which may be a convolutional layer, a dense layer, a long short-term memory (LSTM) layer, a non-linear layer, a pooling layer, a batch normalization layer, or a padding layer). As another example, the first deep learning neural network may include any suitable number of neurons in the various layers (e.g., different layers may have the same or different numbers of neurons from each other). As yet another example, the first deep learning neural network may include any suitable activation function among the various activation functions (e.g., different neurons may have the same or different activation functions from each other) (e.g., softmax, sigmoid, hyperbolic tangent, rectified linear unit). As yet another example, the first deep learning neural network may include any suitable inter-neuron connections or inter-layer connections (e.g., forward connections, skip connections, recurrent connections).

[0037] Regardless of its specific internal architecture, the first deep learning neural network can be configured as a generative text-to-text model. That is, the first deep learning neural network can be configured to receive any suitable text data (in various cases, this text data may or may not be accompanied by any suitable numerical data or any suitable graphical data) as input, and the first deep learning neural network can be configured to produce synthetic text content (e.g., one or more synthetic sentences or sentence fragments) as output, which synthetic text content is semantically or substantially based on such input text data (and, when appropriate, based on accompanying numerical or graphical data).

[0038] Now, in various aspects, the atlas component can electronically store, maintain, control, or otherwise access a design discovery tree associated with a medical imaging scanner. In various instances, the design discovery tree can be a hierarchical structure of text prompts that have known semantic relationships with each other. In various cases, each text prompt in the design discovery tree can be a natural language question asking about the corresponding technical details, features, parts, operations, or other hardware- or software-based aspects of the medical imaging scanner. In various instances, the text prompt can include a variable text field that is configured or intended to be filled with a specific name or identifier of the medical imaging scanner or any of its parts, or with a specific name or identifier of an event that can occur with respect to, be implemented by, or otherwise affect the medical imaging scanner or any of its parts.

[0039] As a non-limiting example, the first text prompt in the design discovery tree can ask "[Insert scanner identifier here] What are all the subsystems?" As another non-limiting example, the first text prompt can have a "has component" relationship with a second text prompt, and the second text prompt can ask "[Insert subsystem identifier here] What are all the constituent components?" As yet another non-limiting example, the second text prompt can have a "has error" relationship with a third text prompt, and the third text prompt can ask "[Insert component identifier here] What are all the possible error codes?" As even another non-limiting example, the second text prompt can have a "has sensor" relationship with a fourth text prompt, and the fourth text prompt can ask "[Insert component identifier here] What are all the sensors?" As yet another non-limiting example, the second text prompt can have a "has fault" relationship with a fifth text prompt, and the fifth text prompt can ask "[Insert component identifier here] What are all the possible fault modes?" As another non-limiting example, the fifth text prompt can have a "has cause" relationship with a sixth text prompt, and the sixth text prompt can ask "[Insert fault mode identifier here] What are all the possible causes?"

[0040] In any case, the graph component can electronically generate a knowledge graph representing multiple design documents by iteratively executing a first deep learning neural network on a design discovery tree. In various aspects, such execution can be facilitated in a RAG manner, and during such execution, the multiple design documents can serve as RAG references or RAG context.

[0041] More specifically, the knowledge graph can initially be empty (e.g., can have no nodes and no edges). In various instances, the graph component can sequentially iterate through the text prompts of the design discovery tree, starting from the most upstream text prompt and continuing in the order of the relational hierarchy until all text prompts in the design discovery tree have been utilized.

[0042] In some cases, prior to beginning such sequential iteration, the graph component may already know (e.g., due to being notified by a user, operator, or technician via a human-machine interface device) the specific name or identifier of a medical imaging scanner. In various aspects, the graph component can modify the design discovery tree by inserting the specific name or identifier of the medical imaging scanner into any variable text field of the design discovery tree that is configured for or intended for the specific name or identifier of the medical imaging scanner.

[0043] Now, for any given text prompt being considered by the graph component, the graph component can insert an empty node shell into the knowledge graph. If the given text prompt has a known relationship with any other text prompt that has already been considered by the graph component in the design discovery tree (which will not occur for the initially first text prompt considered by the graph component), then the graph component can insert into the knowledge graph edges representing those known relationships between the empty node shell and any nodes in the knowledge graph that already respectively represent those already considered text prompts.

[0044] In various aspects, the graph component can search through the multiple design documents to identify one or more design documents that are substantially coherent, relevant, or applicable to the given text prompt. In various aspects, the graph component can facilitate such search via any suitable document search technique (e.g., keyword-based document search technique, embedding-based document search technique, probabilistic retrieval-based document search technique). In various instances, a design document can be considered coherent, relevant, or applicable to the given text prompt if the substantial content of the design document is in some way related to any semantic question being asked by the given text prompt.

[0045] As a non-limiting example, assume the given text prompt asks about a specific subsystem of a medical imaging scanner. In such a case, any service manual paragraph, page, or section that describes that specific subsystem can be considered coherent, relevant, or applicable to the given text prompt.

[0046] As another non - limiting example, assume that a given text prompt instead asks about a specific failure mode of a medical imaging scanner. In this case, any failure report paragraph, page, or section that describes that specific error mode can be considered relevant, related, or applicable to the given text prompt.

[0047] In various aspects, the atlas component can execute a first deep - learning neural network on a given text prompt and one or more relevant design documents, and such execution can cause the first deep - learning neural network to generate synthetic text content. More specifically, the atlas component can concatenate the given text prompt and one or more relevant design documents, and the atlas component can feed this concatenation into the input layer of the first deep - learning neural network. In various cases, this concatenation can complete a forward pass through one or more hidden layers of the first deep - learning neural network. In various instances, this forward pass can cause the output layer of the first deep - learning neural network to compute synthetic text content based on the activations provided by one or more hidden layers.

[0048] In any case, the synthetic text content can be any suitable natural - language text string that can semantically or substantially answer the question asked by the given text prompt. As a non - limiting example, assume that the given text prompt requests the identification of all subsystems of a medical imaging scanner. In this case, the synthetic text content can be one or more sentences or sentence fragments that explicitly identify or name those subsystems. As another non - limiting example, assume that the given text prompt asks for the identification of all possible causes of a specific failure mode of a medical imaging scanner. In this case, the synthetic text content can be one or more sentences or sentence fragments that explicitly identify or name those causes.

[0049] In various aspects, the atlas component can use this synthetic text content to fill an empty node shell. Thus, the node shell can be considered no longer empty.

[0050] In various instances, the atlas component can determine whether the synthetic text content answers a given text prompt by identifying, stating, or otherwise calling out multiple discrete entities associated with a medical imaging scanner (e.g., multiple component parts of the medical imaging scanner, multiple functions that can be implemented by the medical imaging scanner, multiple events that can occur with respect to the medical imaging scanner). In various cases, the atlas component can facilitate such determination via any suitable named entity recognition (NER) technique (e.g., statistical NER, rule-based NER, transformer-based NER). If so, the atlas component can further modify the design discovery tree by: for each discrete entity in the synthetic text content, making a corresponding copy of any branch of the design discovery tree downstream of the given text prompt; and inserting the name or identifier of the discrete entity into an appropriate variable text field in the corresponding copy.

[0051] As a non-limiting example, assume that the given text prompt requests an explicit identification of all subsystems of a medical imaging scanner. Further, assume that the given text prompt has a "has failure" relationship with another text prompt, and that the other text prompt asks "What are all the possible failure modes of [insert subsystem identifier here]?". Still further, assume that the synthetic text content states that "The medical imaging scanner has a gantry subsystem, an X-ray tube subsystem, a patient table subsystem, and a detector subsystem". In this case, the atlas component can determine via NER that the synthetic text content answers the given text prompt by explicitly calling out four discrete entities: the gantry subsystem; the X-ray tube subsystem; the patient table subsystem; and the detector subsystem. Accordingly, the atlas component can make a corresponding copy of the other text prompt (and all other text prompts downstream of the other text prompt) for each of these four discrete entities. This can result in the design discovery tree now having the following: a first copy or version of the other text prompt that asks "What are all the possible failure modes of the gantry subsystem?"; a second copy or version of the other text prompt that asks "What are all the possible failure modes of the X-ray tube subsystem?"; a third copy or version of the other text prompt that asks "What are all the possible failure modes of the patient table subsystem?"; and a fourth copy or version of the other text prompt that asks "What are all the possible failure modes of the detector subsystem?".

[0052] At this point, the atlas component can be considered to have completed the given text prompt and can accordingly iterate to the next text prompt in the design discovery tree.

[0053] In this way, the graph component can incrementally add to or otherwise expand the knowledge graph one node at a time by iteratively executing the first deep learning neural network on the design discovery tree. After the graph component has considered all the nodes of the design discovery tree, the knowledge graph can be considered complete or finished.

[0054] Note that in various instances, it is possible for a technician or engineer to add new text prompts to the design discovery tree, or add edits or entirely new files to multiple design documents. In fact, such new text prompts, edits, or new files can result from ongoing research and development being conducted by the technician or engineer. In such cases, the graph component can again iterate through the design discovery tree (or any suitable portion thereof) to keep the knowledge graph in sync with the research and development efforts of the technician or engineer.

[0055] In any case, the knowledge graph can substantially or semantically represent the various technical details or features of the medical imaging scanner described or conveyed by multiple design documents, and the graph component can generate the knowledge graph by iteratively executing the first deep learning neural network on the design discovery tree in a RAG manner.

[0056] In various embodiments, the access component can electronically receive, retrieve, or otherwise access a natural language query from any suitable source. In various aspects, the natural language query can be a plain text interrogative sentence that asks some substantial question about any suitable technical feature, component, or malfunction symptom of the medical imaging scanner 104. In various instances, the natural language query can be provided by a user or operator of the medical imaging scanner via any suitable human-machine interface device (e.g., keyboard, keypad, touch screen, voice command).

[0057] In various cases, the query component can electronically store, maintain, control, or otherwise access a second deep learning neural network. In various aspects, the second deep learning neural network can exhibit any suitable internal architecture. For example, the second deep learning neural network can include any suitable number of any suitable type of layer (e.g., an input layer, one or more hidden layers, an output layer, and any of these layers can be a convolutional layer, a dense layer, an LSTM layer, a non-linear layer, a pooling layer, a batch normalization layer, or a padding layer). As another example, the second deep learning neural network can include any suitable number of neurons in various layers (e.g., different layers can have the same or different numbers of neurons from each other). As yet another example, the second deep learning neural network can include any suitable activation function among various activation functions (e.g., different neurons can have the same or different activation functions from each other) (e.g., softmax, sigmoid, hyperbolic tangent, rectified linear unit). As yet another example, the second deep learning neural network can include any suitable inter-neuron connections or inter-layer connections (e.g., forward connections, skip connections, recurrent connections).

[0058] Regardless of its specific internal architecture, the second deep learning neural network can be configured to convert an unstructured text query into a structured text query. Thus, the query component can execute the second deep learning neural network on a natural language query, and such execution can cause the second deep learning neural network to produce a structured query as output. More specifically, the query component can feed the natural language query into the input layer of the second deep learning neural network, the natural language query can complete a forward pass through one or more hidden layers of the second deep learning neural network, and the output layer of the second deep learning neural network can calculate the structured query based on the activations provided by one or more hidden layers of the second deep learning neural network. In any case, the structured query can be considered a substantial or semantic version of the natural language query that exhibits a defined or standardized format (e.g., SQL format, Cypher format).

[0059] In various aspects, the query component can identify an electronic answer or response to a natural language question by executing the structured query on a knowledge graph. In various instances, this can be facilitated by any suitable graph query technique (e.g., facilitated by SQL, SPARQL, GraphQL, or Cypher). In some cases, the query component can visually present the electronic answer or response on any suitable computer screen or monitor visible to the user or operator who asked the natural language query. In other cases, the query component can transmit the electronic answer or response to any suitable computing device associated with the user or operator who asked the natural language query. In any case, the query component can be considered to notify the user or operator of the electronic answer or response.

[0060] Note that, in order for the knowledge graphs and the electronic answers or responses described herein to be accurate or reliable, the first deep learning neural network and the second deep learning neural network should undergo training. Accordingly, the computerized tools described herein may include a training component that can facilitate such training in any suitable manner (e.g., in a supervised manner, an unsupervised manner, a reinforcement learning manner).

[0061] The various embodiments described herein can be employed to use hardware or software to solve problems that are highly technical in nature (e.g., facilitating knowledge graph construction via generative artificial intelligence), which are not abstract and cannot be performed by humans as a set of mental acts. Additionally, some of the processes performed can be carried out by a dedicated computer (e.g., a deep learning neural network performing on a design discovery tree) in order to implement the defined actions related to the knowledge graph. For example, such defined actions can include: accessing, by a device operatively coupled to a processor, a plurality of electronic documents associated with the design or manufacture of a medical imaging scanner; and constructing, by the device, a knowledge graph representing the plurality of electronic documents by iteratively performing a generative text-to-text neural network on a design discovery tree associated with the medical imaging scanner. Additionally, such defined actions can include: accessing, by the device, a natural language query regarding the medical imaging scanner; converting, by the device, the natural language query into a structured query via the execution of another neural network; and performing, by the device, the structured query on the knowledge graph, thereby generating an electronic answer to the natural language query. Further still, each element of the design discovery tree can be a text prompt that requests an identification of the corresponding technical design details of the medical imaging scanner, which can have a known relationship with another text prompt of the design discovery tree, and the defined actions can also include: inserting, by the device, a node shell into the knowledge graph; attaching, by the device, an instantiation of the known relationship to the node shell; and populating, by the device, the node shell by performing a generative text-to-text neural network on the text prompt in a RAG manner, where the plurality of design documents are considered RAG references or RAG context.

[0062] Such defined actions are not performed manually by a human. In fact, neither the human mind nor a human with pen and paper can: electronically create a knowledge graph representing technical information of a medical imaging scanner by iteratively performing a generative text-to-text neural network on a design discovery tree in a RAG manner; and electronically answer natural language queries regarding a medical imaging scanner by converting a natural language query to a structured query via another neural network and performing the structured query on the knowledge graph. In fact, a medical imaging scanner, deep learning neural networks, a knowledge graph, and structured queries are all inherently computerized, hardware-based, or software-based constructs that the human mind simply cannot meaningfully implement, train, or perform in any way without a computer. Computerized tools (which can automatically create a knowledge graph via iterative execution of a generative text-to-text neural network and answer questions by converting those questions to structured queries and then performing those structured queries on the knowledge graph) are likewise inherently computerized and cannot be implemented in any sensible, practical, or reasonable way without a computer.

[0063] In addition, the various embodiments described herein can integrate various teachings related to knowledge graph construction via generative artificial intelligence into practical applications. As described above, questions regarding a medical imaging scanner can be answered by performing appropriate queries on a knowledge graph representing technical information of the medical imaging scanner. To facilitate this, the knowledge graph must first be created or constructed. The prior art creates or constructs a knowledge graph via a text-to-graph parser. Unfortunately, as recognized by the inventors of the present invention, a text-to-graph parser can be considered to exhibit poor generality across technical domains. Thus, the prior art requires extensive retraining each time a knowledge graph in a new technical domain is needed. Such extensive retraining can be considered labor-intensive, time-consuming, or otherwise undesirable.

[0064] The various embodiments described herein can solve one or more of these technical problems. In particular, the inventors of the present invention have designed various techniques for constructing a knowledge graph via generative artificial intelligence. Specifically, the inventors of the present invention recognized that generative text-to-text models can exhibit higher generality than text-to-graph parsers. However, the inventors of the present invention also recognized that prior to the teachings described herein, it was not at all clear how to use a generative text-to-text model to construct a knowledge graph. In fact, a generative text-to-text model synthesizes plain text as output, but a knowledge graph is not plain text. Instead, a knowledge graph is a graph data structure with nodes and edges between the nodes. Prior to the work of the inventors of the present invention, automatically converting the synthesized plain text into such a graph data structure without using a text-to-graph parser was a non-trivial task that could not be achieved by the prior art. Despite this non-triviality, the inventors of the present invention have designed the various embodiments described herein. In various aspects, when given a set of design documents describing technical or manufacturing information about a medical imaging scanner, the various embodiments described herein can include: generating a knowledge graph representing the set of electronic documents by iteratively executing a generative text-to-text model on a design discovery tree associated with the medical imaging scanner. In various instances, the design discovery tree can be an ordered hierarchical structure of text prompts that have known semantic relationships with each other. In various cases, each text prompt can be a plain text interrogative sentence that asks about some corresponding technical details or features of the medical imaging scanner. In various aspects, the knowledge graph can initially be empty, and the various embodiments described herein can incrementally expand the knowledge graph by iteratively traversing the design discovery tree in a top-down order. In particular, for any given text prompt in the design discovery tree, the various embodiments described herein can insert an empty node shell (along with any accompanying semantic relationships) into the knowledge graph, and such embodiments can fill the empty node shell by executing a generative text-to-text model on the given text prompt. In various instances, this execution can be facilitated in a RAG manner using the set of design documents as a reference or context. Additionally, this execution can produce synthetic text content that substantially or semantically answers the given text prompt. Thus, the synthetic text content can be inserted into the empty node shell. By iteratively traversing all the text prompts in the design discovery tree in this manner, the knowledge graph can be incrementally constructed (e.g., one node at a time). Therefore, the various embodiments described herein can facilitate knowledge graph construction via generative text-to-text artificial intelligence. Because generative models can exhibit greater generality than text-to-graph parsers (e.g., consider ChatGPT, which can be considered a generative text-to-text model that exhibits high performance across a very wide range of different technical fields), the various embodiments described herein can be considered an improved way of constructing a knowledge graph compared to the prior art.Accordingly, the various embodiments described herein of course constitute tangible and specific technological improvements or technological advantages in the field of knowledge graphs. Accordingly, such embodiments are clearly eligible as useful and practical applications of a computer.

[0065] In addition, the various embodiments described herein can control tangible devices in the real world based on the disclosed teachings. For example, the various embodiments described herein can electronically train and execute a real-world deep learning neural network in order to create a real-world knowledge graph representing technical characteristics or manufacturing information regarding a real-world medical imaging scanner.

[0066] It should be understood that the figures and description herein provide non-limiting examples of the various embodiments and are not necessarily drawn to scale.

[0067] Figure 1 A block diagram of an example non-limiting system 100 that can facilitate knowledge graph construction via generative artificial intelligence, in accordance with one or more embodiments described herein, is illustrated. As shown, a knowledge graph construction system 102 can be electronically integrated with a plurality of design documents 106 that can correspond to a medical imaging scanner 104 via any suitable wired or wireless electronic connection.

[0068] In various embodiments, the medical imaging scanner 104 can be any suitable device, apparatus, or modality for capturing or generating medical images. As a non-limiting example, the medical imaging scanner 104 can be a CT scanner capable of capturing or generating a CT scan pixel array or voxel array. As another non-limiting example, the medical imaging scanner 104 can be an MRI scanner capable of capturing or generating an MRI scan pixel array or voxel array. As yet another non-limiting example, the medical imaging scanner 104 can be an X-ray scanner capable of capturing or generating an X-ray scan pixel array or voxel array. As yet another non-limiting example, the medical imaging scanner 104 can be an ultrasound scanner capable of capturing or generating an ultrasound scan pixel array or voxel array. As yet another non-limiting example, the medical imaging scanner 104 can be a PET scanner capable of capturing or generating a PET scan pixel array or voxel array. As another non-limiting example, the medical imaging scanner 104 can be an NM scanner capable of capturing or generating an NM scan pixel array or voxel array.

[0069] In various aspects, the medical imaging scanner 104 can be deployed, positioned, or otherwise implemented in any suitable clinical operating environment. As a non-limiting example, the medical imaging scanner 104 can be deployed, positioned, or otherwise implemented within any suitable hospital or medical center. As another non-limiting example, the medical imaging scanner 104 can be deployed, positioned, or otherwise implemented within any suitable scientific or medical laboratory. As yet another non-limiting example, the medical imaging scanner 104 can be deployed, positioned, or otherwise implemented within any suitable veterinary center. As even another non-limiting example, the medical imaging scanner 104 can be deployed, positioned, or otherwise implemented within or on any suitable vehicle (e.g., ambulance, cruise ship, airplane).

[0070] In various cases, the medical imaging scanner 104 can include any suitable human-machine interface device through which a user or operator of the medical imaging scanner 104 can manually interact with or control the medical imaging scanner 104. As a non-limiting example, the medical imaging scanner 104 can include any suitable keyboard or keypad that can be pressed by a user or operator. As another non-limiting example, the medical imaging scanner 104 can include any suitable computer mouse that can be dragged or clicked by a user or operator. As even another non-limiting example, the medical imaging scanner 104 can include any suitable touch screen that can be tactually manipulated by a user or operator. As yet another non-limiting example, the medical imaging scanner 104 can include any suitable voice command system that can accept verbal instructions spoken by a user or operator.

[0071] In various scenarios, a plurality of design documents 106 can include n nodes, where n > 1 is any suitable positive integer: design document 106(1) through design document 106(n). In various aspects, each design document among the plurality of design documents 106 can be any suitable electronic data (e.g., one or more scalars, one or more vectors, one or more matrices, one or more tensors, one or more strings, or any suitable combination thereof), which can be related in some way to the technical design or manufacture of the medical imaging scanner 104 (e.g., design document 106(1) can be the first electronic file related to the technical design or manufacture of the medical imaging scanner 104; design document 106(n) can be the nth electronic file related to the technical design or manufacture of the medical imaging scanner 104). In other words, each design document among the plurality of design documents 106 can indicate, specify, convey, describe, explain, show, or otherwise represent in a textual, numerical, or graphical manner: any suitable information regarding how to technically or scientifically design the medical imaging scanner 104 (e.g., different component parts of the medical imaging scanner 104 can be listed or shown and how they work can be described); any suitable information regarding how to manufacture or produce the medical imaging scanner 104 (e.g., what materials the medical imaging scanner 104 is made of can be listed or explained or what production processes are used to manufacture the medical imaging scanner 104); any suitable information regarding how the medical imaging scanner 104 functions (e.g., what different settings or parameters the medical imaging scanner 104 has can be listed or shown and how those different settings can be configured or invoked can be described); any suitable information regarding how the medical imaging scanner 104 is expected or hypothesized to be used or operated (e.g., various operating conditions or use case scenarios that the medical imaging scanner 104 is designed to handle or not designed to handle can be shown or explained); or any suitable information regarding how the medical imaging scanner 104 is expected or hypothesized to be maintained (e.g., various maintenance tasks or troubleshooting steps that can be performed by or on the medical imaging scanner 104 can be listed or shown and when such maintenance tasks or troubleshooting steps should be performed can be described).

[0072] As a non - limiting example, any one of the multiple design documents 106 can be a service manual or guide (or any of its paragraphs, pages, chapters, or other parts) known to correspond to the medical imaging scanner 104 or any of its constituent parts. As another non - limiting example, any one of the multiple design documents 106 can be a bill of materials (or any of its paragraphs, pages, sections, or other parts) known to correspond to the medical imaging scanner 104 or any of its constituent parts. As yet another non - limiting example, any one of the multiple design documents 106 can be a manufacturing report or production order (or any of its paragraphs, pages, sections, or other parts) known to correspond to the medical imaging scanner 104 or any of its constituent parts. As another non - limiting example, any one of the multiple design documents 106 can be a blueprint, schematic, or other image (or any of its pages, sections, or parts) known to show the medical imaging scanner 104 or any of its constituent parts. As yet another non - limiting example, any one of the multiple design documents 106 can be a design failure mode and effects analysis (or any of its paragraphs, pages, chapters, or other parts) known to correspond to the medical imaging scanner 104 or any of its constituent parts.

[0073] In various instances, any one of the multiple design documents 106 can be written or otherwise created by a technician or engineer responsible for designing, developing, prototyping, testing, or producing the medical imaging scanner 104 via any suitable word - processing software (e.g., Microsoft )、via any suitable computer - aided design software (e.g., ) or via any suitable quantitative analysis software (e.g., Microsoft ).

[0074] In some aspects, any one of the multiple design documents 106 can be any suitable electronic data or file that can be created, generated, or otherwise output by the medical imaging scanner 104 itself. As some non - limiting examples, any one of the multiple design documents 106 can be: a pixel array, a voxel array, or a sinogram captured by the medical imaging scanner 104; an error report output by the medical imaging scanner 104; or any time series output or measured by any suitable sensor of the medical imaging scanner 104.

[0075] In any case, it may be desirable to generate a knowledge graph representing the multiple design documents 106 such that questions about the medical imaging scanner 104 can be automatically answered via knowledge - graph query techniques. As described herein, the knowledge - graph construction system 102 can facilitate or achieve such purposes.

[0076] In various embodiments, the knowledge graph construction system 102 may include a processor 108 (e.g., a computer processing unit, a microprocessor) and a non-transitory computer-readable memory 110 that is operatively or operationally or communicatively connected or coupled to the processor 108. The non-transitory computer-readable memory 110 may store computer-executable instructions that, when executed by the processor 108, may cause the processor 108 or other components of the knowledge graph construction system 102 (e.g., the access component 112, the graph component 114, the query component 116) to perform one or more actions. In various embodiments, the non-transitory computer-readable memory 110 may store computer-executable components (e.g., the access component 112, the graph component 114, the query component 116), and the processor 108 may execute the computer-executable components.

[0077] In various embodiments, the knowledge graph construction system 102 may include an access component 112. In various aspects, the access component 112 may electronically access a medical imaging scanner 104 or a plurality of design documents 106. As a non-limiting example, the access component 112 is capable of electronically communicating with the medical imaging scanner 104 in any suitable manner. That is, the access component 112 is capable of electronically transmitting any suitable electronic data to the medical imaging scanner 104, and the medical imaging scanner 104 may similarly transmit any suitable electronic data to the access component 112 in an electronic manner. As another non-limiting example, the access component 112 may electronically retrieve or otherwise electronically obtain a plurality of design documents 106 from any suitable centralized or decentralized data structure (not shown) or from any suitable centralized or decentralized computing device (not shown). In fact, in some cases, the access component 112 may electronically receive a plurality of design documents 106 from the medical imaging scanner 104. In any case, the access component 112 may electronically access the medical imaging scanner 104 or the plurality of design documents 106 such that the access component 112 may serve as a conduit through which other components of the knowledge graph construction system 102 may electronically interact with the medical imaging scanner 104 or with the plurality of design documents 106.

[0078] In various embodiments, the knowledge graph construction system 102 may include a graph component 114. In various aspects, as described herein, the graph component 114 may construct a knowledge graph representing a plurality of design documents 106 by iteratively performing a generative text-to-text model on a design discovery tree associated with the medical imaging scanner 104.

[0079] In various embodiments, the knowledge graph construction system 102 may include a query component 116. In various instances, as described herein, the query component 116 may identify answers to any given query regarding the medical imaging scanner 104 by leveraging the knowledge graph constructed by the graph component 114.

[0080] Figure 2 A block diagram of an example non - limiting system 200 that can facilitate knowledge graph construction via generative artificial intelligence in accordance with one or more embodiments described herein is illustrated. The system includes a deep learning neural network, a design discovery tree, and a knowledge graph. As shown, in some cases, system 200 may include the same components as system 100 and may further include a deep learning neural network 202, a design discovery tree 204, and a knowledge graph 206.

[0081] In various embodiments, the graph component 114 may store, maintain, control, or otherwise access the deep learning neural network 202 electronically. In various instances, the deep learning neural network 202 may have or otherwise exhibit any suitable deep learning internal architecture. For example, the deep learning neural network 202 may have an input layer, one or more hidden layers, and an output layer. In various instances, any of such layers may be coupled together by any suitable inter - neuron or inter - layer connections (such as forward connections, skip connections, or recurrent connections). Additionally, in various cases, any of such layers may be any suitable type of neural network layer having any suitable learnable or trainable internal parameters. For example, any of such input layer, one or more hidden layers, or output layer may be a convolutional layer, and the learnable or trainable parameters thereof may be convolutional kernels. As another example, any of such input layer, one or more hidden layers, or output layer may be a dense layer, and the learnable or trainable parameters thereof may be weight matrices or bias values. As yet another example, any of such input layer, one or more hidden layers, or output layer may be a batch normalization layer, and the learnable or trainable parameters thereof may be shift factors or scale factors. As even another example, any of such input layer, one or more hidden layers, or output layer may be an LSTM layer, and the learnable or trainable parameters thereof may be input state weight matrices or hidden state weight matrices. Further still, in various cases, any of such layers may be any suitable type of neural network layer having any suitable fixed or non - trainable internal parameters. For example, any of such input layer, one or more hidden layers, or output layer may be a non - linear layer, a padding layer, a pooling layer, or a concatenation layer.

[0082] Regardless of the specific internal architecture implemented within the deep learning neural network 202 (e.g., the specific number, type, or organization of layers), the deep learning neural network 202 can be configured as a generative text-to-text model. In other words, the deep learning neural network 202 can be configured to receive text data (which may be accompanied by any suitable numerical or graphical data) as input and produce synthetic text data (e.g., synthetic natural language sentences or sentence fragments) as output, where such synthetic text data is semantically or substantially based on the input text data.

[0083] In various aspects, the atlas component 114 can electronically store, electronically maintain, electronically control, and / or otherwise electronically access the design discovery tree 204. In various instances, the design discovery tree 204 can be a hierarchical structure of text prompts, where such text prompts can have known semantic relationships with each other, and where each of such text prompts can be a plain text or natural language question about the corresponding technical details of the medical imaging scanner 104.

[0084] In various cases, the knowledge graph 206 can be any suitable graph data structure with nodes and edges, which can represent any technical information or manufacturing information of the medical imaging scanner 104 described or conveyed by the multiple design documents 106. In various aspects, the atlas component 114 can electronically generate, construct, or otherwise create the knowledge graph 206 by iteratively executing the deep learning neural network 202 on the design discovery tree 204, where such execution can be facilitated in a RAG manner using the multiple design documents 106 as references or context. With respect to Figures 3 to 12 Non-limiting aspects are described.

[0085] Figure 3 An example non-limiting block diagram 300 of the design discovery tree 204 in accordance with one or more embodiments described herein is illustrated.

[0086] As described above, the design discovery tree 204 can be an ordered hierarchical structure of text prompts that have known semantic relationships with each other. In other words, the design discovery tree 204 can be considered a directed acyclic graph whose nodes are text prompts and whose edges are semantic relationships. In various aspects, the design discovery tree 204 can include any suitable number of text prompts arranged in any suitable manner via any suitable number of semantic relationships. In various instances, each text prompt of the design discovery tree 204 can be a plain text or natural language string that requests or commands an identification of corresponding technical details of the medical imaging scanner 104. Non-limiting examples of such technical details can include: the constituent hardware or software components that make up the medical imaging scanner 104; the functions that can be performed by such constituent hardware or software components; the possible error codes that can be output or thrown by the medical imaging scanner 104 or any of its parts; the possible measurement results that can be captured or measured by a sensor of the medical imaging scanner 104 or any of its parts; the possible diagnostic tools that can be implemented by or with respect to the medical imaging scanner 104 or any suitable part thereof; the possible failure modes that the medical imaging scanner 104 or any of its parts can experience; or the possible causes of such failure modes. Because the design discovery tree 204 can be a directed acyclic graph, the design discovery tree 204 can be considered to have one most upstream text prompt such that all other text prompts of the design discovery tree 204 are downstream of that most upstream text prompt (e.g., branch directly or indirectly from that most upstream text prompt). Additionally, any text prompt of the design discovery tree can include a variable text field into which the name or identifier of the medical imaging scanner 104, the name or identifier of any part of the medical imaging scanner 104, or the name or identifier of any event, function, or other entity associated with the medical imaging scanner 104 can be inserted. In particular, the most upstream text prompt of the design discovery tree 204 (also referred to as the root text prompt) can include a variable text field that is configured or intended to be populated with the specific name or identifier of the medical imaging scanner 104. In contrast, any given text prompt that is not the most upstream or root text prompt can instead include a variable text field that is configured or intended to be populated with the specific name or identifier of any one or more entities that answer any text prompt immediately upstream of that given text prompt.

[0087] In Figure 3In a non-limiting example, the design discovery tree 204 can include a text prompt 302 as its root. In various aspects, the text prompt 302 can correspond to or otherwise relate to design details 304, where the design details 304 can be any suitable general or common feature, aspect, property, characteristic, or category that is related in some way to the technical design, technical function, technical manufacture, technical maintenance, or technical troubleshooting of the medical imaging scanner 104. Thus, the text prompt 302 can be any suitable interrogative or imperative sentence or sentence fragment that requests or commands the identification of the design details 304. As a non-limiting example, assume that the design details 304 are a subsystem category. In this case, the text prompt 302 can be: the following interrogative sentence "[What are all the subsystems of [insert scanner identifier here]?]"; or the following imperative sentence "[Identify all the subsystems of [insert scanner identifier here]]".

[0088] In various aspects, the text prompt 302 can have a known semantic relationship 306 with a text prompt 308 such that the text prompt 308 can be considered to be directly downstream of the text prompt 302. In various instances, the text prompt 308 can correspond to or otherwise relate to design details 310, where the design details 310 can be any suitable general or common feature, aspect, property, characteristic, or category that is related in some way to the technical design, technical function, technical manufacture, technical maintenance, or technical troubleshooting of the medical imaging scanner 104. In various cases, the design details 310 can be different from or otherwise not the same as the design details 304. Additionally, in various aspects, the design details 310 can depend on or otherwise be related to the known semantic relationship 306. In any case, the text prompt 302 can be any suitable interrogative or imperative sentence or sentence fragment that requests or commands the identification of the design details 310.

[0089] As a non-limiting example, as described above, assume that the design details 304 are a subsystem category. Additionally, assume that the known semantic relationship 306 is a "has components" relationship, and assume that the design details 310 are a parts category. In this case, the text prompt 308 can be: the following interrogative sentence "[What are all the component parts of [insert subsystem identifier here]?]"; or the following imperative sentence "[Identify all the component parts of [insert subsystem identifier here]]".

[0090] In various aspects, the text prompt 308 may have a known semantic relationship 312 with the text prompt 314 such that the text prompt 314 can be considered to be directly downstream of the text prompt 308 and indirectly downstream of the text prompt 302. In various instances, the text prompt 314 may correspond to or otherwise be related to design details 316, where the design details 316 can be any suitable general or generic feature, aspect, property, characteristic, or category that is in some way related to the technical design, technical function, technical manufacture, technical maintenance, or technical troubleshooting of the medical imaging scanner 104. In various cases, the design details 316 may be different from or otherwise not the same as the design details 304 and the design details 310. Additionally, in various aspects, the design details 316 may depend on or otherwise be related to the known semantic relationship 312. In any case, the text prompt 314 can be any suitable interrogative or imperative sentence or sentence fragment that requests or commands the identification of the design details 316.

[0091] As a non-limiting example, as described above, assume that the design details 310 are a part category. Additionally, assume that the known semantic relationship 312 is a "has error" relationship, and assume that the design details 316 are an error code category. In this case, the text prompt 314 can be: the following interrogative sentence "What are all the possible error codes that can be output by [insert component part identifier here]?"; or the following imperative sentence "Identify all the possible error codes that can be output by [insert component part identifier here]."

[0092] As another non-limiting example, as described above, assume that the design details 310 are a part category. Additionally, assume that the known semantic relationship 312 is a "has sensor" relationship, and assume that the design details 316 are a sensor category. In this case, the text prompt 314 can be: the following interrogative sentence "What are all the sensors that [insert component part identifier here] is equipped with?"; or the following imperative sentence "Identify all the sensors that [insert component part identifier here] is equipped with."

[0093] As yet another non-limiting example, as described above, assume that the design details 310 are a part category. Additionally, assume that the known semantic relationship 312 is a "has diagnostic test" relationship, and assume that the design details 316 are a diagnostic test category. In this case, the text prompt 314 can be: the following interrogative sentence "What are all the possible diagnostic tests that can be performed on or by [insert component part identifier here]?"; or the following imperative sentence "Identify all the possible diagnostic tests that can be performed on or by [insert component part identifier here]."

[0094] In various aspects, the text prompt 308 may have a known semantic relationship 318 with the text prompt 320 such that the text prompt 320 can be considered to be directly downstream of the text prompt 308 and indirectly downstream of the text prompt 302. In various instances, the text prompt 320 may correspond to or otherwise relate to design details 322, where the design details 322 can be any suitable general or generic feature, aspect, property, characteristic, or category that is in some way related to the technical design, technical function, technical manufacture, technical maintenance, or technical troubleshooting of the medical imaging scanner 104. In various cases, the design details 322 may be different from or otherwise distinct from the design details 304, the design details 310, and the design details 316. Additionally, in various aspects, the design details 322 may depend on or otherwise be related to the known semantic relationship 318. In any case, the text prompt 320 can be any suitable interrogative or imperative sentence or sentence fragment that requests or commands the identification of the design details 322.

[0095] As a non - limiting example, as described above, assume that the design detail 310 is a part category. Additionally, assume that the known semantic relationship 318 is a "has failure modes" relationship, and assume that the design detail 322 is a failure mode category. In this case, the text prompt 320 can be: the following interrogative sentence "[What are all the possible failure modes that [insert component part identifier here] can experience?]"; or the following imperative sentence "Identify all the possible failure modes that [insert component part identifier here] can experience".

[0096] In various aspects, the text prompt 320 may have a known semantic relationship 324 with the text prompt 326 such that the text prompt 326 can be considered to be directly downstream of the text prompt 320 and indirectly downstream of the text prompt 308 and the text prompt 302. In various instances, the text prompt 326 may correspond to or otherwise relate to design details 328, where the design details 328 can be any suitable general or generic feature, aspect, property, characteristic, or category that is in some way related to the technical design, technical function, technical manufacture, technical maintenance, or technical troubleshooting of the medical imaging scanner 104. In various cases, the design details 328 may be different from or otherwise distinct from the design details 304, the design details 310, the design details 316, and the design details 322. Additionally, in various aspects, the design details 328 may depend on or otherwise be related to the known semantic relationship 324. In any case, the text prompt 326 can be any suitable interrogative or imperative sentence or sentence fragment that requests or commands the identification of the design details 328.

[0097] As a non - limiting example, as described above, assume that design detail 322 is a failure mode category. Further, assume that the known semantic relation 324 is a "has cause" relation, and assume that design detail 328 is a cause category. In this case, the text prompt 320 can be: the following interrogative sentence "[What are all the possible causes of [insert failure mode identifier here]?]"; or the following imperative sentence "Identify all the possible causes of [insert failure mode identifier here]".

[0098] In various cases, as described above, the text prompts in the design discovery tree 204 can include or contain variable text fields (e.g., indicated by [insert identifier here] in the above example). In various aspects, the atlas component 114 can modify the design discovery tree 204 by inserting appropriate names or identifiers into those variable text fields. In some instances, this can involve making copies or duplicates of various text prompts and inserting such names or identifiers into the variable text fields of those copies or duplicates. Indeed, as described herein, the atlas component 114 can utilize the deep - learning neural network 202 to synthesize responses or otherwise generate corresponding text content for each text prompt in the design discovery tree 204. Thus, for any given text prompt, the atlas component 114 can synthesize a response or generate the text content that responds to that given text prompt. If the synthesized text content explicitly includes or names multiple discrete entities, corresponding copies of any text prompts downstream of that given text prompt can be made for each of those multiple discrete entities, and the names or identifiers of those multiple discrete entities can be individually inserted into the appropriate variable text fields of those corresponding copies. Thus, in Figure 3 the non - limiting example shown, the text prompt 308 (and the known semantic relation 306) can be repeated or duplicated as needed based on any text content synthesized by the atlas component 114 for the text prompt 302. Similarly, the text prompt 314 (and the known semantic relation 312) can be repeated or duplicated as needed based on any text content synthesized by the atlas component 114 for the text prompt 308 (e.g., based on any text content synthesized for each copy of the text prompt 308). Similarly, the text prompt 320 (and the known semantic relation 318) can be repeated or duplicated as needed based on any text content synthesized by the atlas component 114 for the text prompt 308 (e.g., based on any text content synthesized for each copy of the text prompt 308). Further, the text prompt 326 can be repeated or duplicated as needed based on any text content synthesized by the atlas component 114 for the text prompt 320 (e.g., based on any text content synthesized for each copy of the text prompt 320). This is explained in more detail with respect to Figure 12 this point.

[0099] In any case, the atlas component 114 can electronically execute the deep learning neural network 202 in a RAG manner for each text prompt of the design discovery tree 204 in a top - down order or sequence, where multiple design documents 106 are used as RAG references or context. This enables the atlas component 114 to synthesize corresponding text content substantially responsive to each text prompt in the text prompts of the design discovery tree 204. In various aspects, each such synthesized text content can be used as or otherwise regarded as a corresponding node of the knowledge graph 206, and such nodes can be coupled with known semantic relationships of the design discovery tree 204 when appropriate. With respect to Figures 4 to 12 Non - limiting aspects are described.

[0100] Figures 4 to 12 Illustrative non - limiting block diagrams 400, 500, 600, 700, 800, 900, 1000, 1100, and 1200 according to one or more embodiments described herein show how a knowledge graph 206 can be incrementally constructed via generative artificial intelligence. In particular, the atlas component 114 can iterate through the design discovery tree 204 in an upstream - to - downstream direction starting from the most upstream (or root) text prompt. In Figure 3 a non - limiting example of, this can mean that the atlas component 114 can start at text prompt 302 and then proceed to text prompt 308 and thereafter proceed to the remaining text prompts in a breadth - first or depth - first manner. Figures 4 to 7 Shows how the atlas component 114 can consider text prompt 302. Figures 8 to 11 Shows how the atlas component 114 can then consider text prompt 308. Figure 12 Relates to how a text prompt can be copied based on any text content synthesized upstream of the text prompt.

[0101] First, consider Figure 4 . In various aspects, the atlas component 114 can electronically create the knowledge graph 206 so that it is initially empty. In other words, the knowledge graph 206 may initially not have nodes and edges. As shown, when the atlas component 114 considers text prompt 302, in various instances, the atlas component 114 can insert a node shell 402 into the knowledge graph 206. The node shell 402 can be considered a node representing the text prompt 302 that does not yet have any substantial content. In other words, the node shell 402 can initially be empty.

[0102] Now, consider Figure 5。In various embodiments, the atlas component 114 may electronically search through the plurality of design documents 106 to identify which of the plurality of design documents 106 are substantially relevant to the text prompt 302. In various aspects, the atlas component 114 may implement this via any suitable document search technique.

[0103] As a non-limiting example, the atlas component 114 may apply any suitable keyword-based document search technique to the plurality of design documents 106. In such cases, the atlas component 114 may identify (e.g., via named entity identification) one or more keywords included or recited in the text prompt 302, and the atlas component 114 may search the plurality of design documents 106 for any (if any) design documents that also include or recite that one or more keywords. If any given design document includes or recites that one or more keywords, then that given design document may be considered to be substantially relevant or otherwise related to the text prompt 302. For example, if the text prompt 302 asks "What are all the failure modes of the rack motor?", then any design document that recites the keywords "failure" and "rack motor" may be considered to be substantially relevant to the text prompt 302.

[0104] As another non-limiting example, the atlas component 114 may apply any suitable embedding-based document search technique to the plurality of design documents 106. In such cases, the atlas component 114 may generate (e.g., via a deep learning autoencoder trained in an unsupervised manner) a first embedding (e.g., a latent vector representation) of the text prompt 302, the atlas component 114 may similarly generate or otherwise access a second embedding of the corresponding design documents in the plurality of design documents 106, and the atlas component 114 may determine which of those second embeddings (if any) are sufficiently similar (e.g., in terms of Euclidean distance or cosine similarity) to the first embedding. If the second embedding of any given design document is within any suitable threshold level of similarity to the first embedding of the text prompt 302, then that given design document may be considered to be substantially relevant or otherwise related to the text prompt 302 (e.g., considered to include text, numerical, or graphical data that is substantially relevant or otherwise related to the text prompt).

[0105] As yet another non-limiting example, the atlas component 114 may apply any suitable probability-based document search technique to the plurality of design documents 106. For example, the atlas component 114 may utilize a term frequency-inverse document frequency (TF-IDF) search technique. As another example, the atlas component 114 may utilize a best match 25 search technique.

[0106] Regardless of the type of document search technology implemented, the atlas component 114 can search through multiple design documents 106 to identify one or more design documents that are substantially relevant to the text prompt 302. In various aspects, such substantially relevant documents may be considered or otherwise referred to as one or more relevant design documents 502. In various instances, as shown, the one or more relevant design documents 502 may include m documents, where m ≤ n is any suitable positive integer: relevant design document 502(1) to relevant design document 502(m). In some cases, the atlas component 114 may sort the multiple design documents 106 in order of their substantial relevance to the text prompt 302 (e.g., in a keyword manner, in an embedded manner, or in a probabilistic manner), and any m documents with the highest ranking among these documents may be considered the one or more relevant design documents 502.

[0107] Now, consider Figure 6 . In various aspects, the atlas component 114 can electronically execute the deep learning neural network 202 on the text prompt 302 and on the one or more relevant design documents 502. In various instances, such execution can cause the deep learning neural network 202 to generate synthetic text content 602. More specifically, the atlas component 114 can concatenate the text prompt 302 with the one or more relevant design documents 502 to produce a concatenation. In various cases, the atlas component 114 can feed or route the concatenation to the input layer of the deep learning neural network 202. In various aspects, the concatenation can complete a forward pass through one or more hidden layers of the deep learning neural network 202. In various instances, the output layer of the deep learning neural network 202 can estimate or calculate the synthetic text content 602 based on the activation map or feature map generated by the one or more hidden layers.

[0108] In any case, the synthetic text content 602 can be one or more declarative sentences or sentence fragments that are substantially or semantically responsive to any question or command conveyed by the text prompt 302 as generated by the deep learning neural network 202. More specifically, as described above, the text prompt 302 can request or command the identification of design details 304 of the medical imaging scanner 104, and the one or more relevant design documents 502 can be considered to describe or explain (possibly in a long, multi-paragraph, or multi-page manner) information that is substantially relevant to the text prompt 302 and thus to the design details 304. In various aspects, the deep learning neural network 202 can be considered to succinctly distill this information into a plain text or natural language sentence or sentence fragment that clearly identifies the design details 304 of the medical imaging scanner 104, and such a sentence or sentence fragment can be considered the synthetic text content 602.

[0109] Now, consider Figure 7。In various aspects, the graph component 114 may insert the synthetic text content 602 into the node shell 402. In other words, the node shell 402 may initially be empty, but may now be populated with the synthetic text content 602. In various instances, the graph component 114 may complete the text prompt 302 and may thus advance to the text prompt 308.

[0110] Consider Figure 8 。In various aspects, the graph component 114 may electronically insert the node shell 802 into the knowledge graph 206. In various instances, the node shell 802 may be considered to be a node representing the text prompt 308 that does not yet have any substantial content. In other words, the node shell 802 may initially be empty. Since the text prompt 308 has a known semantic relationship 306 with the text prompt 302 and since the node shell 402 represents the text prompt 302, the graph component 114 may further insert an edge representing the known semantic relationship 306 between the node shell 402 and the node shell 802.

[0111] Now, consider Figure 9 。In various embodiments, the graph component 114 may electronically search through the plurality of design documents 106 to identify which of the plurality of design documents 106 are substantially relevant to the text prompt 308. As described above, the graph component 114 may achieve this via any suitable document search technique (e.g., keyword-based search, embedding-based search, probability-based search). Regardless of the type of document search technique implemented, the graph component 114 may search through the plurality of design documents 106 to identify one or more design documents that are substantially relevant to the text prompt 308. In various aspects, such substantially relevant documents may be considered or otherwise referred to as one or more relevant design documents 902. In various instances, as shown, the one or more relevant design documents 902 may include p documents, where p ≤ n is any suitable positive integer: relevant design document 902(1) through relevant design document 902(p). In some cases, the graph component 114 may sort the plurality of design documents 106 in order of their substantial relevance to the text prompt 308 (e.g., keyword-wise, embedding-wise, or probability-wise), and any p documents that are highest ranked among these documents may be considered the one or more relevant design documents 902. In some cases, p may be equal to m. However, in other cases, p may not be equal to m.

[0112] Now, consider Figure 10。In various aspects, the atlas component 114 may electronically execute the deep learning neural network 202 on the text prompt 308 and on one or more coherent design documents 902. In various instances, such execution may cause the deep learning neural network 202 to generate synthetic text content 1002. More specifically, the atlas component 114 may concatenate the text prompt 308 with one or more coherent design documents 902 to produce a concatenation. In various cases, the atlas component 114 may feed or route the concatenation to the input layer of the deep learning neural network 202. In various aspects, the concatenation may complete a forward pass through one or more hidden layers of the deep learning neural network 202. In various instances, the output layer of the deep learning neural network 202 may estimate or compute the synthetic text content 1002 based on the activation maps or feature maps generated by one or more hidden layers.

[0113] In any case, the synthetic text content 1002 may be one or more declarative sentences or sentence fragments that are substantially or semantically responsive to any question or command conveyed by the text prompt 308 as generated by the deep learning neural network 202. More specifically, and as described above, the text prompt 308 may request or command the identification of design details 310 of the medical imaging scanner 104; one or more coherent design documents 902 may be considered to describe or explain (possibly in a verbose, multi-paragraph, or multi-page manner) information that is substantially coherent with the text prompt 308 and thus with the design details 310; the deep learning neural network 202 may be considered to succinctly distill such information into a plain text or natural language sentence or sentence fragment that clearly identifies the design details 310 of the medical imaging scanner 104; and such sentence or sentence fragment may be considered the synthetic text content 1002.

[0114] Now, consider Figure 11 。In various aspects, the atlas component 114 may insert the synthetic text content 1002 into the node housing 802. In other words, the node housing 802 may initially be empty but may now be populated with the synthetic text content 1002. In various instances, the atlas component 114 may complete the text prompt 308 and may thus proceed to the next or subsequent text prompt.

[0115] Now, consider Figure 12 。In various aspects, as described above, each text prompt in the design discovery tree 204 may include a variable text field, and the atlas component 114 may modify or alter the design discovery tree 204 by making appropriate copies of such each text prompt and populating the variable text fields of those copies with appropriate entity names or identifiers. Figure 12 An illustrative non-limiting example of how this may occur with respect to the text prompt 308 is provided.

[0116] In fact, assume that the synthetic text content 602 explicitly contains, identifies, or names t different entities in response to the text prompt 302, where t > 1 is any suitable positive integer (note that each of the t different entities can be considered to belong to any technical category indicated by the design details 304). In various aspects, the graph component 114 can determine the t different entities identified or named by the synthetic text content 602 by applying any suitable NER technique to the synthetic text content 602. In some cases, the graph component 114 can perform rule-based NER on the synthetic text content 602, such as provided by the OpenNLP software platform. In other cases, the graph component 114 can perform statistic-based NER on the synthetic text content 602, such as provided by the SpaCy software platform. In even other cases, the graph component 114 can perform transformer-based NER on the synthetic text content 602.

[0117] In any case, since the synthetic text content 602 can answer the text prompt 302 with the explicit identification of t different entities, the graph component 114 can make a total of t copies of all text prompts that directly or indirectly branch from the text prompt 302 in the design discovery tree 204, and the graph component 114 can separately insert the names or identifiers of those t different entities into the appropriate variable text fields of those t copies.

[0118] In particular, the text prompt 308 can include a variable text field into which each of the t different entities can be separately inserted. Thus, the graph component 114 can cause there to be a total of t different versions of the text prompt 308 in the design discovery tree 204: a first version of the text prompt 308 whose variable text field is filled with the name or identifier of the first different entity to a t-th version of the text prompt 308 whose variable text field is filled with the name or identifier of the t-th different entity. In various aspects, the graph component 114 can process each of those t versions of the text prompt 308 as described above, such that a coherence search and node content synthesis based on the design details 310 can be separately performed for each of those t different entities.

[0119] For example, with respect to the first version of text prompt 308, the graph component 114 may: insert the initially empty node shell 802(1) that represents the first version of text prompt 308 into the knowledge graph 206; insert an edge representing the known semantic relationship 306 between node shell 402 and node shell 802(1); identify one or more relevant design documents for the first version of text prompt 308; synthesize the synthetic text content 1002(1) for the first version of text prompt 308 via the deep learning neural network 202 based on the one or more relevant design documents; and may insert the synthetic text content 1002(1) into node shell 802(1).

[0120] Similarly, with respect to the t-th version of text prompt 308, the graph component 114 may: insert the initially empty node shell 802(t) that represents the t-th version of text prompt 308 into the knowledge graph 206; insert an edge representing the known semantic relationship 306 between node shell 402 and node shell 802(t); identify one or more relevant design documents for the t-th version of text prompt 308; synthesize the synthetic text content 1002(t) for the t-th version of text prompt 308 via the deep learning neural network 202 based on the one or more relevant design documents; and may insert the synthetic text content 1002(t) into node shell 802(t).

[0121] In this way, the graph component 114 may modify or extend the design discovery tree 204 based on how many different entities are identified, recounted, or called out in any text context synthesized for any given text prompt, and thus modify or extend the knowledge graph 206.

[0122] To help clarify various aspects, consider the following non-limiting example. Suppose the medical imaging scanner 104 is an MRI scanner with "MRI12345" as an identifier (e.g., this could be the model number or serial number of the medical imaging scanner 104). Additionally, suppose the text prompt 302 asks "[Insert scanner identifier here] What are all the subsystems?" In various cases, the graph component 114 may insert the identifier "MRI12345" into the variable text field of text prompt 302, such that text prompt 302 becomes "What are all the subsystems of MRI12345?"

[0123] Now, assume that one or more coherent design documents 502 are paragraphs, pages, or diagrams from a service manual of a medical imaging scanner 104 that describe, explain, or illustrate: the gantry of an MRI 12345 scanner having "G123" as a model or serial number; the X-ray tube of the MRI 12345 scanner having "X123" as a model or serial number; and the patient table of the MRI 12345 scanner having "T123" as a model or serial number. In this case, the synthetic text content 602 can be a declarative sentence stating that "the MRI 12345 scanner has a gantry G123 subsystem, an X-ray tube X123 subsystem, and a patient table T123 subsystem".

[0124] In various aspects, the atlas component 114 can determine via NER that the synthetic text content 602 answers the text prompt 302 in this case by explicitly reciting, specifying, or naming the following three different entities (e.g., t = 3): "the gantry G123 subsystem"; "the X-ray tube X123 subsystem"; and "the patient table T123 subsystem".

[0125] Now, assume that the text prompt 308 asks "What are all the parts of [insert subsystem identifier here]?". In various instances, the atlas component 114 can create a corresponding version of the text prompt 308 for each of the three different entities recited in the synthetic text content 602. In particular, the first version of the text prompt 308 can ask "What are all the parts of the gantry G123 subsystem?"; the second version of the text prompt 308 can ask "What are all the parts of the X-ray tube X123 subsystem?"; and the third version of the text prompt 308 can ask "What are all the parts of the patient table T123 subsystem?".

[0126] Note that the first version of the text prompt 308 can be accompanied by corresponding versions or copies of all other text prompts that branch directly or indirectly from the text prompt 308 (e.g., the first version of the text prompt 314 can branch directly from the first version of the text prompt 308; the first version of the text prompt 320 can branch directly from the first version of the text prompt 308; and the first version of the text prompt 326 can branch directly from the first version of the text prompt 320).

[0127] Similarly, note that the second version of the text prompt 308 can be accompanied by corresponding versions or copies of all other text prompts that branch directly or indirectly from the text prompt 308 (e.g., the second version of the text prompt 314 can branch directly from the second version of the text prompt 308; the second version of the text prompt 320 can branch directly from the second version of the text prompt 308; and the second version of the text prompt 326 can branch directly from the second version of the text prompt 320).

[0128] Similarly, it should be noted that the third version of the text prompt 308 may be accompanied by corresponding versions or copies of all other text prompts that branch directly or indirectly from the text prompt 308 (e.g., the third version of the text prompt 314 may branch directly from the third version of the text prompt 308; the third version of the text prompt 320 may branch directly from the third version of the text prompt 308; and the third version of the text prompt 326 may branch directly from the third version of the text prompt 320).

[0129] In any case, the atlas component 114 can synthesize text content for each of the three versions of the text prompt 308, as described above, and can insert the synthesized text content into the corresponding node shells in the knowledge graph 206.

[0130] In this way, the atlas component 114 can incrementally create, construct, or expand the knowledge graph 206 one or more nodes at a time by iteratively performing the deep learning neural network 202 on the design discovery tree 204 in a RAG manner. After all the text prompts in the design discovery tree 204 have been used to synthesize the node content of the knowledge graph 206, the knowledge graph 206 can be considered complete or finished (e.g., at least until new text prompts are added to the design discovery tree 204 or new documents are added to the plurality of design documents 106).

[0131] Figures 13 to 15 Flowcharts illustrate example non - limiting computer - implemented methods 1300, 1400, and 1500 for facilitating knowledge graph construction via generative artificial intelligence according to one or more embodiments described herein. In various cases, the knowledge graph construction system 102 can facilitate the computer - implemented methods 1300, 1400, and 1500.

[0132] First, consider Figure 13 . In various embodiments, the action 1302 can include: accessing, by a device (e.g., 112) operatively coupled to a processor (e.g., 108), a plurality of design documents (e.g., 106) related to a medical imaging scanner (e.g., 104).

[0133] In various aspects, the action 1304 can include: accessing, by the device (e.g., via 114), a design discovery tree (e.g., 204) associated with the medical imaging scanner. In various cases, the design discovery tree can be a hierarchical structure of text prompts that have known relationships with each other and that query general or common technical details of the medical imaging scanner (e.g., "What are the subsystems of [insert scanner identifier here]?"; "What are the sensors of [insert subsystem identifier here]?"; "What are the possible failure modes of [insert subsystem identifier here]?").

[0134] In various examples, action 1306 may include: creating, by a device (e.g., via 114), a knowledge graph (e.g., 206) that is initially empty.

[0135] In various cases, action 1308 may include: determining, by a device (e.g., via 114), whether all text cues in the design discovery tree have been considered? If so, the computer-implemented method 1300 may end, meaning the knowledge graph may be considered complete or finalized. If not, the computer-implemented method 1300 may instead proceed to action 1310.

[0136] In various aspects, action 1310 may include: selecting, by a device (e.g., via 114), the most upstream text cue (e.g., 302) in the design discovery tree that has not yet been considered.

[0137] In various examples, action 1312 may include: inserting, by a device (e.g., 114), an initially empty node shell (e.g., 402) representing the selected text cue into the knowledge graph. In various cases, the computer-implemented method 1300 may then proceed to action 1402 of the computer-implemented method 1400.

[0138] Now, consider Figure 14 . In various embodiments, action 1402 may include: determining, by a device (e.g., via 114), whether the selected text cue has a known relationship (e.g., 306) with other text cues in the design discovery tree that have already been considered. If so, the computer-implemented method 1400 may proceed to action 1404. If not, the computer-implemented method 1400 may instead proceed to action 1406.

[0139] In various aspects, action 1404 may include: inserting, by a device (e.g., via 114), the known relationship between the node shell and any nodes in the knowledge graph representing those other text cues into the knowledge graph. In various cases, the computer-implemented method 1400 may then proceed back to action 1406.

[0140] In various examples, action 1406 may include: identifying, by a device (e.g., via 114), one or more design documents (e.g., 502) from a plurality of design documents that are substantially relevant to the selected text cue.

[0141] In various cases, operation 1408 may include: synthesizing text content (e.g., 602) by a device (e.g., via 114) via execution of a generative artificial intelligence (AI) model (e.g., 202), the text content answering a selected text prompt and being based on one or more substantially coherent design documents. In various aspects, computer-implemented method 1400 may proceed to operation 1502 of computer-implemented method 1500.

[0142] Now, consider Figure 15 . In various embodiments, operation 1502 may include: determining, by a device (e.g., via 114), whether the synthesized text content answers the selected text prompt by calling out or reciting a plurality of discrete entities associated with a medical imaging scanner (e.g., t entities identified via NER). If not, computer-implemented method 1500 may return to operation 1308 of computer-implemented method 1300. If so, computer-implemented method 1500 may proceed to operation 1504.

[0143] In various aspects, operation 1504 may include: modifying the design discovery tree by a device (e.g., via 114) by making a respective copy of any branch of the design discovery tree downstream of the selected text prompt for the plurality of discrete entities and inserting the name or identifier of the plurality of discrete entities individually into respective variable text fields of those respective copies. In various cases, computer-implemented method 1500 may then return to operation 1308 of computer-implemented method 1300.

[0144] In any case, the atlas component 114 may build the knowledge graph 206 by leveraging the deep learning neural network 202 and the design discovery tree 204. After such building, the knowledge graph building system 102 may utilize the knowledge graph 206 to answer any technical questions that may exist for a user or operator of the medical imaging scanner 104, such as regarding Figures 16 to 17 as described.

[0145] Figure 16 Illustrated is a block diagram of an example non-limiting system 1600 that may facilitate knowledge graph table construction via generative artificial intelligence in accordance with one or more embodiments described herein, the system including natural language queries, another deep learning neural network, structured queries, and answers. As shown, in some cases, system 1600 may include the same components as system 200 and may further include natural language query 1602, deep learning neural network 1604, structured query 1606, and answer 1608.

[0146] In various embodiments, the access component 112 may electronically receive, electronically retrieve, or otherwise electronically access the natural language query 1602 from any suitable data structure or source. In various aspects, the natural language query 1602 may be any suitable plain text question asking about some technical aspect or feature of the medical imaging scanner 104. As a non-limiting example, the natural language query 1602 may ask why the medical imaging scanner 104 produces image artifacts. As another non-limiting example, the natural language query 1602 may ask what maintenance should be performed on the medical imaging scanner 104 or how often such maintenance should be performed. In various instances, the natural language query 1602 may be provided or input by a user or operator of the medical imaging scanner 104 via any suitable human-machine interface device (e.g., via a keyboard, keypad, touch screen, or voice control system of the medical imaging scanner 104).

[0147] In various aspects, the query component 116 may electronically store, electronically maintain, electronically control, or otherwise electronically access the deep learning neural network 1604. In various instances, the deep learning neural network 1604 may have or otherwise exhibit any suitable deep learning internal architecture. For example, the deep learning neural network 1604 may have an input layer, one or more hidden layers, and an output layer. In various instances, any of such layers may be coupled together by any suitable inter-neuron or inter-layer connections (such as forward connections, skip connections, or recurrent connections). Additionally, in various cases, any of such layers may be any suitable type of neural network layer having any suitable learnable or trainable internal parameters. For example, any of such input layer, one or more hidden layers, or output layer may be a convolutional layer, and the learnable or trainable parameters of the convolutional layer may be convolutional kernels. As another example, any of such input layer, one or more hidden layers, or output layer may be a dense layer, and the learnable or trainable parameters of the dense layer may be weight matrices or bias values. As yet another example, any of such input layer, one or more hidden layers, or output layer may be a batch normalization layer, and the learnable or trainable parameters of the batch normalization layer may be shift factors or scale factors. As yet another example, any of such input layer, one or more hidden layers, or output layer may be an LSTM layer, and the learnable or trainable parameters of the LSTM layer may be input state weight matrices or hidden state weight matrices. Further still, in various cases, any of such layers may be any suitable type of neural network layer having any suitable fixed or non-trainable internal parameters. For example, any of such input layer, one or more hidden layers, or output layer may be a non-linear layer, a padding layer, a pooling layer, or a concatenation layer.

[0148] Regardless of the specific internal architecture implemented within the deep learning neural network 1604 (e.g., the specific number, type, or organization of layers), the deep learning neural network 1604 can be configured as an unstructured to structured query transformer or translator. That is, the deep learning neural network 1604 can be configured to receive an unstructured query as input and produce a corresponding structured query as output. Thus, by leveraging the deep learning neural network 1604, the query component 116 can convert the natural language query 1602 into a structured query 1606. In various cases, the query component 116 can then generate an answer 1608 by executing the structured query 1606 against the knowledge graph 206. With respect to Figure 17 non-limiting aspects are described.

[0149] Figure 17 FIG. 1700 is an example non-limiting block diagram illustrating how a natural language query can be converted into a structured query 1606 and then a structured query can be executed against the knowledge graph 206, in accordance with one or more embodiments described herein.

[0150] In various embodiments, as described above, the natural language query 1602 can be a plain text interrogative sentence that asks about some technical features of the medical imaging scanner 104 and is provided by a user or operator of the medical imaging scanner 104. In various aspects, the query component 116 can electronically execute the deep learning neural network 1604 on the natural language query 1602. In various instances, such execution can cause the deep learning neural network 1604 to produce a structured query 1606. More specifically, the query component 116 can feed or route the natural language query 1602 to the input layer of the deep learning neural network 1604. In various aspects, the natural language query 1602 can complete a forward pass through one or more hidden layers of the deep learning neural network 1604. In various instances, the output layer of the deep learning neural network 1604 can estimate or calculate the structured query 1606 based on the activation map or feature map generated by one or more hidden layers.

[0151] In any case, the structured query 1606 can be considered a strictly formatted version of the natural language query 1602 suitable for graph execution. As a non-limiting example, the deep learning neural network 1604 can be configured to convert plain text into SQL format. In this case, the structured query 1606 can be considered a semantically equivalent version of the natural language query 1602 written according to SQL syntax or otherwise formatted. As another non-limiting example, the deep learning neural network 1604 can be configured to convert plain text into SPARQL format. In this case, the structured query 1606 can be considered a semantically equivalent version of the natural language query 1602 written according to SPARQL syntax or otherwise formatted. As yet another non-limiting example, the deep learning neural network 1604 can be configured to convert plain text into GraphQL format. In this case, the structured query 1606 can be considered a semantically equivalent version of the natural language query 1602 written according to GraphQL syntax or otherwise formatted. As yet another non-limiting example, the deep learning neural network 1604 can be configured to convert plain text into Cypher format. In this case, the structured query 1606 can be considered a semantically equivalent version of the natural language query 1602 written according to Cypher syntax or otherwise formatted.

[0152] In various aspects, the query component 116 can electronically execute the structured query 1606 against the knowledge graph 206. In various instances, such execution can be facilitated in any suitable manner (e.g., in SQL manner, in SPARQL manner, in GraphQL manner, in Cypher manner). In any case, such execution can produce an answer 1608. In various aspects, the answer 1608 can be any suitable electronic data (e.g., one or more scalars, one or more vectors, one or more matrices, one or more tensors, one or more strings or any suitable combination thereof), which can be considered to represent any information queried or requested in the structured query 1606 and thus in the natural language query 1602. As a non-limiting example, again consider the above instance where the natural language query 1602 can query why the medical imaging scanner 104 produces image artifacts. In this case, the answer 1608 can recite or identify one or more potential causes of the imaging artifacts of the medical imaging scanner 104 enumerated in the knowledge graph 206. As another non-limiting example, again consider the above instance where the natural language query 1602 can query what maintenance should be performed on the medical imaging scanner 104 or how often such maintenance should be performed. In this case, the answer 1608 can recite or identify one or more recommended maintenance tasks or maintenance schedules of the medical imaging scanner 104 enumerated in the knowledge graph 206.

[0153] In various aspects, the query component 116 can electronically present the response 1608 on any suitable electronic display (e.g., a screen) of the medical imaging scanner 104 such that the response 1608 can be seen by a user or operator of the medical imaging scanner 104. In other instances, the query component 116 can electronically transmit the response 1608 to any other suitable computing device (e.g., a smartphone) associated with the user or operator such that the user or operator can be aware of the response 1608.

[0154] Thus, as described herein, the knowledge graph construction system 102 can incrementally build the knowledge graph 206 by iteratively performing the deep learning neural network 202 on the design discovery tree 204 in a RAG manner, and the knowledge graph construction system 102 can then utilize the knowledge graph 206 to answer real-world questions that a real-world user or operator may have regarding the medical imaging scanner 104.

[0155] For the knowledge graph construction system 102 to operate accurately, correctly, or reliably, the deep learning neural network 202 and the deep learning neural network 1604 can first undergo training, as described with respect to Figures 18 to 19 described.

[0156] Figure 18 FIG. illustrates a block diagram of an example non-limiting system 1800 that can facilitate knowledge graph construction via generative artificial intelligence in accordance with one or more embodiments described herein, the system including a training component. As shown, in some cases, the system 1800 can include the same components as the system 1600 and can also include a training component 1802. In various instances, the training component 1802 can use any suitable training paradigm to train the deep learning neural network 202 or the deep learning neural network 1604. In some cases, such training can be facilitated in a supervised manner, as described with respect to Figure 19 described.

[0157] Figure 19 FIG. illustrates an example non-limiting block diagram 1900 showing how a deep learning neural network can be trained in accordance with one or more embodiments described herein.

[0158] In various aspects, prior to beginning training, the training component 1802 can initialize the trainable internal parameters (e.g., convolutional kernels, weight matrices, bias values) of the deep learning neural network 202 (or the deep learning neural network 1604) in any suitable manner (e.g., via random initialization).

[0159] In various embodiments, there may be a training input 1902 and a ground truth annotation 1904. When training a deep learning neural network 202 is desired, the training input 1902 may be a training text prompt (or a concatenation of the training text prompt and any suitable number of training relevant design documents), and the ground truth annotation 1904 may be correct or accurate synthetic text content known or believed to correspond to the training input 1902. Conversely, when training a deep learning neural network 1604 is desired, the training input 1902 may be a training natural language query, and the ground truth annotation 1904 may be a correct or accurate structured query known or believed to correspond to the training input 1902.

[0160] In any case, the training component 1802 may execute the deep learning neural network 202 (or the deep learning neural network 1604) on the training input 1902 such that the deep learning neural network 202 (or the deep learning neural network 1604) produces an output 1906. More specifically, in some cases, the training component 1802 may feed or route the training input 1902 to the input layer of the deep learning neural network 202 (or the deep learning neural network 1604), the training input 1902 may complete a forward pass through one or more hidden layers of the deep learning neural network 202 (or the deep learning neural network 1604), and the output layer of the deep learning neural network 202 (or the deep learning neural network 1604) may compute the output 1906 based on the activation map or feature map provided by one or more hidden layers of the deep learning neural network 202 (or the deep learning neural network 1604).

[0161] Note that the format, size, or dimension of the output 1906 may be determined by the number, arrangement, size, or other characteristics of the neurons, convolutional kernels, or other internal parameters of the output layer (or any other layer) of the deep learning neural network 202 (or the deep learning neural network 1604). Thus, the output 1906 can be forced to have any desired format, size, or dimension by adding, removing, or otherwise adjusting the characteristics of the output layer (or any other layer) of the deep learning neural network 202 (or the deep learning neural network 1604).

[0162] In various aspects, if the output 1906 is generated by the deep learning neural network 202, the output 1906 can be considered as the predicted or inferred text content that the deep learning neural network 202 has synthesized based on the training input 1902. On the other hand, if the output 1906 is generated by the deep learning neural network 1604, the output 1906 can be considered as the predicted or inferred structured query that the deep learning neural network 1604 believes should correspond to the training input 1902. In any case, the ground truth annotation 1904 can be considered as any correct or accurate result (e.g., correct or accurate synthesized text content, correct or accurate structured query) that is known or considered to correspond to the training input 1902. Note that if the deep learning neural network 202 (or the deep learning neural network 1604) has hardly undergone training so far, the output 1906 may be highly inaccurate. In other words, the output 1906 can be significantly different from the ground truth annotation 1904.

[0163] In various aspects, the training component 1802 can compute the error (e.g., mean absolute error (MAE), mean squared error (MSE), cross-entropy error) between the output 1906 and the ground truth annotation 1904. In various instances, the training component 1802 can incrementally update the trainable internal parameters of the deep learning neural network 202 (or the deep learning neural network 1604) based on the computed error via backpropagation (e.g., stochastic gradient descent).

[0164] In various cases, this execution and update procedure can be repeated for any suitable number of input-annotation pairs. This can ultimately cause the trainable internal parameters of the deep learning neural network 202 (or the deep learning neural network 1604) to be iteratively optimized for accurately generating synthesized text content (or structured query). In various aspects, the training component 1802 can utilize any suitable training batch size, any suitable error / loss function, or any suitable training termination criterion.

[0165] Although the disclosure herein mainly describes the deep learning neural network 202 and the deep learning neural network 1604 as being trained in a supervised manner, this is merely a non-limiting example for ease of explanation and illustration. In various embodiments, any other suitable training paradigm (such as unsupervised training or reinforcement learning) can be used to train the deep learning neural network 202 and the deep learning neural network 1604.

[0166] Figure 20 A flowchart of an example non-limiting computer-implemented method 2000 that can facilitate knowledge graph construction via generative artificial intelligence according to one or more embodiments described herein is illustrated. In various cases, the knowledge graph construction system 102 can facilitate the computer-implemented method 2000.

[0167] In various embodiments, operation 2002 may include a device (e.g., via 112) operatively coupled to a processor (e.g., 108) accessing a plurality of electronic documents (e.g., 106) associated with the design or manufacture of a medical imaging scanner (e.g., 104).

[0168] In various aspects, operation 2004 may include a device (e.g., via 114) constructing a knowledge graph (e.g., 206) representative of the plurality of electronic documents by iteratively performing a generative text-to-text neural network (e.g., 202) on a design discovery tree (e.g., 204) associated with a medical imaging scanner.

[0169] In various instances, operation 2006 may include a device (e.g., via 112) accessing a natural language query (e.g., 1602) regarding a medical imaging scanner.

[0170] In various cases, operation 2008 may include a device (e.g., via 116) converting the natural language query into a structured query (e.g., 1606) via the execution of another neural network (e.g., 1604).

[0171] In various aspects, operation 2010 may include a device (e.g., via 116) performing a structured query on the knowledge graph, thereby generating an electronic answer (e.g., 1608) to the natural language query.

[0172] Although not explicitly shown in Figure 20 for each element of the design discovery tree, the element may be a text prompt (e.g., 308) that requests an identification of corresponding technical design details (e.g., 310) of a medical imaging scanner.

[0173] Although not explicitly shown in Figure 20 the text prompt may have a known relationship (e.g., 306) with another text prompt (e.g., 302) of the design discovery tree.

[0174] Although not explicitly shown in Figure 20Although not explicitly shown, the computer-implemented method 2000 may include: inserting a node shell (e.g., 802) into the knowledge graph by the device (e.g., via 114); attaching an instantiation of a known relationship to the node shell by the device (e.g., via 114); and filling the node shell by the device (e.g., via 114) by performing a generative text-to-text neural network on the text prompt in a retrieval-augmented generation manner. In various cases, performing a generative text-to-text neural network on the text prompt in a retrieval-augmented generation manner may include: identifying, by the device (e.g., via 114), one or more first electronic documents (e.g., 902) relevant to the text prompt from a plurality of electronic documents; concatenating the text prompt with the one or more first electronic documents by the device (e.g., via 114), thereby producing a concatenation; and performing a generative text-to-text neural network on the concatenation by the device (e.g., via 114), such that the generative text-to-text neural network synthesizes text content (e.g., 1002) describing the corresponding technical design details of the medical imaging scanner. In various cases, the device (e.g., via 114) may fill the node shell with the text content.

[0175] Although in Figure 20 not explicitly shown, the plurality of electronic documents may include: blueprints or schematics associated with the medical imaging scanner; or failure analysis reports associated with the medical imaging scanner.

[0176] Various embodiments have been described herein with respect to the construction of a knowledge graph containing technical information about a medical imaging scanner. However, these are merely non-limiting examples. In various cases, the various embodiments described herein may be applied or extrapolated to construct a knowledge graph of any suitable machine (e.g., not limited to constructing a knowledge graph of a medical imaging scanner).

[0177] In fact, various embodiments may relate to a computer program product for facilitating knowledge graph construction via generative artificial intelligence. In various aspects, the computer program product may include a non-transitory computer-readable memory (e.g., 110) having program instructions embodied therewith. In various instances, the program instructions may be executable by a processor (e.g., 108) to cause the processor to: access a plurality of electronic documents (e.g., 106) associated with the design or manufacture of a machine (e.g., 104); and construct a knowledge graph (e.g., 206) representing the plurality of electronic documents by iteratively executing a generative text-to-text neural network (e.g., 202) on a design discovery tree (e.g., 204) associated with the machine. In various aspects, the program instructions may further be executable to cause the processor to: access a natural language query (e.g., 1602) regarding the machine; convert the natural language query to a structured query (e.g., 1606) via the execution of another neural network (e.g., 1604); and execute the structured query on the knowledge graph, thereby generating an electronic answer (e.g., 1608) to the natural language query. In various instances, for each element of the design discovery tree, the element may be a text prompt (e.g., 308) that requests an identification of a corresponding technical design detail (e.g., 310) of the machine, and the text prompt may have a known relationship (e.g., 306) with another text prompt of the design discovery tree. In various cases, the program instructions may further be executable to cause the processor to: insert a node shell (e.g., 802) into the knowledge graph; attach an instantiation of a known relationship to the node shell; and populate the node shell by executing a generative text-to-text neural network on the text prompt in a retrieval-augmented generation manner.

[0178] In various examples, machine learning algorithms or models can be implemented in any suitable manner to facilitate any suitable aspects described herein. To facilitate some of the machine learning aspects of the above-described machine learning aspects of the various implementations, consider the following discussion of artificial intelligence (AI). The various implementations described herein can employ artificial intelligence to facilitate the automation of one or more features or functionalities. These components can employ various AI-based schemes to perform the various implementations / examples disclosed herein. To provide or facilitate the numerous determinations (e.g., determine, detect, infer, estimate, predict, prognosis, estimate, derive, forecast, detect, calculate) described herein, the components described herein can examine all or a subset of the data to which they are granted access and can provide reasoning about or determine the state of a system or environment from a set of observations captured via events or data. For example, determinations can be used to identify a particular context or action, or a probability distribution of a state can be generated. These determinations can be probabilistic; that is, the calculation of the probability distribution of the state of interest is based on the consideration of data and events. Determination can also refer to techniques for composing higher-level events from a set of events or data.

[0179] Such determinations can result in the construction of new events or actions from a set of observed events or stored event data, regardless of whether the events are temporally close and regardless of whether the events and data are from one or more events and data sources. The components disclosed herein can employ various classification (explicitly trained (e.g., via training data) and implicitly trained (e.g., via observed behavior, preferences, historical information, receiving extrinsic information, etc.)) schemes or systems (e.g., support vector machines, neural networks, expert systems, Bayesian belief networks, fuzzy logic, data fusion engines, etc.) in conjunction with performing automatic or determined actions related to the claimed subject matter. Thus, classification schemes or systems can be used to automatically learn and perform a variety of functions, actions, or determinations.

[0180] A classifier can map an input attribute vector z = (z1, z2, z3, z4, z n ) to a confidence that the input belongs to a certain class, e.g., according to f(z) = confidence(class). Such classification can employ probability- or statistics-based analysis (e.g., analyzing utility and cost considerations) to determine the action to be automatically performed. A support vector machine (SVM) can be an example of a classifier that can be used. An SVM operates by finding a hyperplane in the space of possible inputs, where the hyperplane attempts to separate triggering criteria from non-triggering events. Intuitively, this makes the classification correct for testing data that is close to but different from the training data. Other directed and undirected model classification methods include, for example, naive Bayes, Bayesian networks, decision trees, neural networks, fuzzy logic models, or any of the probabilistic classification models that can provide different independent modes. Classification as used herein also includes statistical regression for developing priority models.

[0181] To provide additional context to the various embodiments described herein, Figure 21 and the following discussion is intended to provide a brief general description of a suitable computing environment 2100 in which various embodiments of the embodiments described herein may be implemented. Although the embodiments have been described above in the general context of computer-executable instructions that may run on one or more computers, those skilled in the art will recognize that these embodiments may also be implemented in combination with other program modules or as a combination of hardware and software.

[0182] Generally, program modules include routines, programs, components, data structures, etc. that perform particular tasks or implement particular abstract data types. In addition, those skilled in the art will appreciate that the methods of the present invention may be practiced with other computer system configurations, including single-processor or multi-processor computer systems, minicomputers, mainframe computers, Internet of Things (IoT) devices, distributed computing systems, and personal computers, handheld computing devices, microprocessor-based or programmable consumer electronics, etc., each of which is operatively coupled to one or more associated devices.

[0183] The illustrated embodiments of the embodiments herein may also be practiced in a distributed computing environment where particular tasks are performed by remote processing devices linked through a communications network. In a distributed computing environment, program modules may be located in local and remote memory storage devices.

[0184] Computing devices generally include a variety of media, which may include computer-readable storage media, machine-readable storage media, or communication media, where the use of these two terms is different from each other herein, as described below. Computer-readable storage media or machine-readable storage media can be any available storage media accessible by a computer and includes volatile and non-volatile media, removable and non-removable media. By way of example and not limitation, computer-readable storage media or machine-readable storage media may be implemented in conjunction with any method or technology for storing information such as computer-readable or machine-readable instructions, program modules, structured data, or unstructured data.

[0185] A computer-readable storage medium may include, but is not limited to, random access memory (RAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), flash memory or other memory technologies, compact disc read only memory (CD-ROM), digital versatile disc (DVD), Blu-ray disc (BD) or other optical disc storage devices, magnetic tape cartridges, magnetic tape, disk storage devices or other magnetic storage devices, solid state drives or other solid state storage devices, or other tangible or non-transitory media that can be used to store the desired information. In this regard, the terms "tangible" or "non-transitory" as applied to storage devices, memory or computer-readable media herein shall be understood to exclude only propagating transitory signals per se as a modifier and not to foreclose rights in all standard storage devices, memory or computer-readable media that are not only propagating transitory signals per se.

[0186] A computer-readable storage medium may be accessed by one or more local or remote computing devices, for example, via an access request, query, or other data retrieval protocol, to implement various operations with respect to the information stored by the medium.

[0187] A communication medium typically includes computer-readable instructions, data structures, program modules, or other structured or unstructured data in a data signal, which may be, for example, a modulated data signal such as a carrier wave or other transmission mechanism, and includes any information delivery or transmission medium. The term "modulated data signal" or "signal" refers to a signal that sets or changes one or more of its characteristics to encode information in one or more signals. By way of example and not limitation, a communication medium includes wired media (such as a wired network or direct wired connection) and wireless media (such as acoustic, RF, infrared, and other wireless media).

[0188] Referring again to Figure 21 , an example environment 2100 for various implementations for implementing the aspects described herein includes a computer 2102, which includes a processing unit 2104, a system memory 2106, and a system bus 2108. The system bus 2108 couples system components, including but not limited to the system memory 2106, to the processing unit 2104. The processing unit 2104 may be any of a variety of commercially available processors. Dual microprocessors and other multiprocessor architectures may also be used as the processing unit 2104.

[0189] The system bus 2108 can be any one of several types of bus structures that can be further interconnected to a memory bus (with or without a memory controller), a peripheral bus, and a local bus using any one of a variety of commercially available bus architectures. The system memory 2106 includes a ROM 2110 and a RAM 2112. The basic input / output system (BIOS) can be stored in non-volatile memory (such as ROM, erasable programmable read-only memory (EPROM), EEPROM), where the BIOS contains basic routines that help transfer information between elements within the computer 2102 during startup, for example. The RAM 2112 can also include high-speed RAM, such as static RAM for caching data.

[0190] The computer 2102 also includes an internal hard disk drive (HDD) 2114 (e.g., EIDE, SATA), one or more external storage devices 2116 (e.g., a magnetic floppy disk drive (FDD) 2116, a memory stick or flash drive reader, a memory card reader, etc.), and a drive 2120 (e.g., such as a solid-state drive, an optical disc drive) that can read from or write to a disk 2122 (such as a CD-ROM disc, a DVD, a BD, etc.). Alternatively, in the case of a solid-state drive, the disk 2122 will not be included unless separate. Although the internal HDD 2114 is illustrated as being within the computer 2102, the internal HDD 2114 can also be configured for use external to a suitable infrastructure (not shown). Additionally, although not shown in the environment 2100, a solid-state drive (SSD) can be used as a supplement or alternative to the HDD 2114. The HDD 2114, the external storage device 2116, and the drive 2120 can be connected to the system bus 2108 via an HDD interface 2124, an external storage interface 2126, and a drive interface 2128, respectively. The interface 2124 for external drive implementations can include at least one or both of the Universal Serial Bus (USB) and Institute of Electrical and Electronics Engineers (IEEE) 1394 interface technologies. Other external drive connection technologies are within the contemplation of the embodiments described herein.

[0191] The drives and their associated computer-readable storage media provide non-volatile storage of data, data structures, computer-executable instructions, and the like. For the computer 2102, the drives and storage media are adapted to store any data in a suitable digital format. Although the above description of computer-readable storage media refers to the corresponding types of storage devices, those skilled in the art should understand that other types of storage media that are computer-readable (whether currently existing or to be developed in the future) can also be used in the exemplary operating environment, and furthermore, any such storage media can contain computer-executable instructions for performing the methods described herein.

[0192] Multiple program modules may be stored in the drive and RAM 2112, including an operating system 2130, one or more application programs 2132, other program modules 2134, and program data 2136. All or part of the operating system, application programs, modules, or data may also be cached in the RAM 2112. The systems and methods described herein may be implemented using a variety of commercially available operating systems or combinations of operating systems.

[0193] The computer 2102 may optionally include emulation technology. For example, a hypervisor (not shown) or other intermediary may emulate the hardware environment for the operating system 2130, and the emulated hardware may optionally be different from Figure 21 that illustrated. In such an implementation, the operating system 2130 may be included in one of a plurality of virtual machines (VMs) hosted at the computer 2102. Additionally, the operating system 2130 may provide a runtime environment to the application programs 2132, such as a Java runtime environment or a.NET framework. The runtime environment is a consistent execution environment that allows the application programs 2132 to run on any operating system that includes the runtime environment. Similarly, the operating system 2130 may support containers, and the application programs 2132 may be in the form of containers that are lightweight, independent, executable software packages that include, for example, the code of the application program, the runtime, system tools, system libraries, and settings.

[0194] Furthermore, the computer 2102 may be enabled using a security module, such as a Trusted Platform Module (TPM). For example, in the case of a TPM, the boot component is hashed in the next boot component and waits for the result to match a security value before loading the next boot component. This process may occur at any layer in the code execution stack of the computer 2102, such as applied to the application program execution level or the operating system (OS) kernel level, thereby achieving security at any code execution level.

[0195] A user may input commands and information into the computer 2102 through one or more wired / wireless input devices (e.g., keyboard 2138, touch screen 2140, and pointing devices such as mouse 2142). Other input devices (not shown) may include a microphone, an infrared (IR) remote control, a radio frequency (RF) remote control or other remote controls, a joystick, a virtual reality controller or virtual reality headset, a gamepad, a stylus, an image input device (e.g., a camera), a gesture sensor input device, a visual motion sensor input device, an emotion or face detection device, a biometric input device (e.g., a fingerprint or iris scanner), etc. These input devices and other input devices are often connected to the processing unit 2104 through an input device interface 2144, and this input device interface may be coupled to the system bus 2108, but these input devices and other input devices may be connected through other interfaces (such as a parallel port, an IEEE 1394 serial port, a game port, a USB port, an IR interface, interfaces, etc.).

[0196] A monitor 2146 or other type of display device may also be connected to the system bus 2108 via an interface (such as a video adapter 2148). In addition to the monitor 2146, a computer typically also includes other peripheral output devices (not shown), such as speakers, printers, etc.

[0197] The computer 2102 may operate in a networked environment using a logical connection to one or more remote computers (such as remote computer 2150) via wired or wireless communication. The remote computer 2150 may be a workstation, a server computer, a router, a personal computer, a portable computer, a microprocessor-based entertainment appliance, a peer device, or other common network nodes, and generally includes many or all of the elements described with respect to the computer 2102, but for the sake of brevity, only the memory / storage device 2152 is illustrated. The depicted logical connection includes a wired / wireless connection to a local area network (LAN) 2154 or a larger network (e.g., a wide area network (WAN) 2156). Such LAN and WAN networking environments are common in offices and companies and facilitate enterprise-wide computer networks (such as intranets), all of which can be connected to a global communication network (e.g., the Internet).

[0198] When used in a LAN networking environment, the computer 2102 may be connected to the local network 2154 through a wired or wireless communication network interface or adapter 2158. The adapter 2158 may facilitate wired or wireless communication with the LAN 2154, and the LAN may also include a wireless access point (AP) disposed thereon for communicating with the adapter 2158 in wireless mode.

[0199] When used in a WAN networking environment, computer 2102 may include a modem 2160, or may be connected to a communication server on WAN 2156 via other components for establishing communication through the WAN 2156 (such as via the Internet). The modem 2160, which can be an internal or external device and a wired or wireless device, can be connected to the system bus 2108 via the input device interface 2144. In a networking environment, program modules depicted relative to computer 2102 or portions thereof can be stored in the remote memory / storage device 2152. It should be understood that the network connections shown are examples, and other components for establishing a communication link between computers can be used.

[0200] When used in a LAN or WAN networking environment, in addition to or as an alternative to the external storage device 2116 described above, computer 2102 can access a cloud storage system or other network-based storage systems, such as but not limited to network virtual machines that provide one or more aspects of information storage or processing. Generally speaking, the connection between computer 2102 and the cloud storage system can be established, for example, by adapter 2158 or modem 2160 via LAN 2154 or WAN 2156, respectively. When connecting computer 2102 to an associated cloud storage system, the external storage interface 2126 can manage the storage provided by the cloud storage system with the help of adapter 2158 or modem 2160, just like other types of external storage devices. For example, the external storage interface 2126 can be configured to provide access to cloud storage sources as if those cloud storage sources were physically connected to computer 2102.

[0201] Computer 2102 may be capable of operating to communicate with any wireless device or entity operatively set up for wireless communication, such as a printer, scanner, desktop computer or portable computer, portable data assistant, communication satellite, any equipment or location associated with a wirelessly detectable tag (such as a kiosk, newsstand, store shelf, etc.), and a telephone. This can include Wi-Fi and wireless technologies. Thus, the communication can be of a predefined structure like a conventional network, or merely an ad hoc communication between at least two devices.

[0202] Figure 22FIG. 2200 is a schematic block diagram of an example computing environment 2200 with which the disclosed subject matter may interact. The example computing environment 2200 includes one or more clients 2210. The clients 2210 can be hardware or software (e.g., threads, processes, computing devices). The example computing environment 2200 also includes one or more servers 2230. The servers 2230 can also be hardware or software (e.g., threads, processes, computing devices). For example, the servers 2230 can house threads to perform transformations by adopting one or more embodiments as described herein. A possible communication between the clients 2210 and the servers 2230 can be in the form of data packets suitable for transfer between two or more computer processes. The example computing environment 2200 includes a communication framework 2250 that can be used to facilitate communication between the clients 2210 and the servers 2230. The clients 2210 are operatively connected to one or more client data repositories 2220 that can be used to store information local to the clients 2210. Similarly, the servers 2230 are operatively connected to one or more server data repositories 2240 that can be used to store information local to the servers 2230.

[0203] The various embodiments can be a system, a method, an apparatus, or a computer program product at any possible technical detail integration level. The computer program product can include a computer-readable storage medium (or media) having computer-readable program instructions thereon for causing a processor to implement aspects of the various embodiments. The computer-readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. The computer-readable storage medium can be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of the computer-readable storage medium can further include the following: a portable computer floppy disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device such as a punch card or raised structures in grooves recorded with instructions, and any suitable combination of the foregoing items. As used herein, the computer-readable storage medium should not be construed to be a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., an optical pulse through an optical fiber cable), or an electrical signal transmitted through a wire.

[0204] The computer-readable program instructions described herein can be downloaded to a corresponding computing / processing device from a computer-readable storage medium or downloaded to an external computer or external storage device via a network (e.g., the Internet, a local area network, a wide area network, or a wireless network). The network can include copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers, or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium within the corresponding computing / processing device. The computer-readable program instructions for carrying out operations of the various implementations can be assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state-setting data, configuration data for integrated circuits, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk, C++, etc., and procedural programming languages such as the "C" programming language or similar programming languages. The computer-readable program instructions can be executed entirely on the user's computer, partially on the user's computer, executed as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can establish a connection with an external computer (e.g., through the Internet using an Internet service provider). In some implementations, an electronic circuit, including, for example, a programmable logic circuit, a field-programmable gate array (FPGA), or a programmable logic array (PLA), can execute the computer-readable program instructions by utilizing the state information of the computer-readable program instructions to personalize the electronic circuit for various aspects.

[0205] Aspects described herein are illustrated by flowcharts or block diagrams of methods, apparatus (systems), and computer program products according to various embodiments. It should be understood that each block of the flowchart illustrations or block diagrams, and combinations of blocks in the flowchart illustrations or block diagrams, can be implemented by computer-readable program instructions. These computer-readable program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in one or more blocks of the flowchart or block diagram. These computer-readable program instructions may also be stored in a computer-readable storage medium that can direct a computer, programmable data processing apparatus, or other device to function in a particular manner, such that the computer-readable storage medium having instructions stored therein comprises an article of manufacture including instructions for implementing aspects of the functions / acts specified in one or more blocks of the flowchart or block diagram. The computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational acts to be performed on the computer, other programmable apparatus, or other device to produce a computer-implemented process, such that the instructions which execute on the computer, other programmable apparatus, or other device implement the functions / acts specified in one or more blocks of the flowchart or block diagram.

[0206] The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments. In this regard, each block in the flowchart illustrations or block diagrams may represent a module, segment, or portion of instructions, which includes one or more executable instructions for implementing the specified logical function. In some alternative implementations, the functions noted in the blocks may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowchart illustrations, and combinations of blocks in the block diagrams or flowchart illustrations, can be implemented by a system based on dedicated hardware that performs the specified functions or acts or a combination of dedicated hardware and computer instructions.

[0207] Although the subject matter has been described above in the general context of computer-executable instructions of a computer program product running on one or more computers, those skilled in the art will recognize that the present disclosure may also be implemented in whole or in part in conjunction with other program modules. Generally, program modules include routines, programs, components, data structures, etc. that perform particular tasks or implement particular abstract data types. In addition, those skilled in the art should recognize that various aspects may be practiced with other computer system configurations, including single-processor or multi-processor computer systems, minicomputers, mainframe computers, and the like, as well as computer, hand-held computing devices (e.g., PDAs, cellular phones), microprocessor-based or programmable consumer or industrial electronic products, and the like. The illustrated aspects may also be practiced in a distributed computing environment where tasks are performed by remote processing devices linked through a communications network. However, some, if not all, aspects of the present disclosure may be practiced on stand-alone computers. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.

[0208] As used in this application, the terms "component", "system", "platform", "interface", etc. may refer to or may include a computer-related entity or an entity related to an operating machine having one or more specific functionalities. Entities disclosed herein may be hardware, a combination of hardware and software, software, or software in execution. For example, a component may be, but is not limited to, a program running on a processor, a processor, an object, an executable, an execution thread, a program, or a computer. By way of illustration, both an application running on a server and the server can be a component. One or more components may reside within a process or execution thread, and a component may be located on one computer or distributed between two or more computers. As another example, a corresponding component may execute in accordance with various computer-readable media having various data structures stored thereon. Components may communicate via local or remote processes, such as in accordance with signals having one or more data packets (e.g., data from one component that interacts with another component in a local system, a distributed system, or a network such as the Internet). As another example, a component may be a device having specific functionality provided by mechanical parts operated by an electrical or electronic circuit that is operated by a software or firmware application executed by a processor. In such cases, the processor may be internal or external to the device and may execute at least a portion of the software or firmware application. As yet another example, a component may be a device that provides specific functionality through electronic components rather than mechanical parts, where the electronic components may include a processor or other components for executing software or firmware that at least partially imparts functionality to the electronic components. In one aspect, a component may be emulated, for example, within a cloud computing system via a virtual machine.

[0209] In addition, the term "or" is intended to mean an inclusive "or" rather than an exclusive "or". That is, unless otherwise specified or clear from the context, "X employs A or B" is intended to mean any natural inclusive permutation. That is, if X employs A; X employs B; or X employs both A and B, then "X employs A or B" is satisfied in any of the foregoing instances. As used herein, the term "and / or" is intended to have the same meaning as "or". In addition, unless otherwise specified or clear from the context as being directed to the singular form, the articles "a" and "an" used in this specification and the drawings generally should be understood to mean "one or more". As used herein, the terms "example" or "exemplary" are used to mean serving as an example, instance, or illustration. For the avoidance of doubt, the subject matter disclosed herein is not limited by such examples. Additionally, any aspect or design described herein as "example" or "exemplary" should not necessarily be understood as being more preferred or advantageous than other aspects or designs, nor does it mean excluding equivalent exemplary structures and techniques known to those of ordinary skill in the art.

[0210] The disclosure herein describes non-limiting examples. For ease of description or explanation, when discussing various examples, each part of the disclosure herein uses the terms "each", "every", or "all". Such uses of the terms "each", "every", or "all" are non-limiting. In other words, when the disclosure herein provides a description applicable to "each", "every", or "all" of some particular objects or components in some particular objects or components, it should be understood that this is a non-limiting example, and it should also be understood that in various examples, there may be cases where such description applies to less than "each", "every", or "all" of the particular objects or components.

[0211] As used in this specification, the term "processor" can generally refer to any computing processing unit or device, including but not limited to a single-core processor; a single processor with software multithreaded execution capabilities; a multi-core processor; a multi-core processor with software multithreaded execution capabilities; a multi-core processor with hardware multithreaded technology; a parallel platform; and a parallel platform with distributed shared memory. Additionally, a processor can refer to an integrated circuit, an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a field-programmable gate array (FPGA), a programmable logic controller (PLC), a complex programmable logic device (CPLD), discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. Furthermore, a processor can utilize nanoscale architectures (such as but not limited to molecule- and quantum dot-based transistors, switches, and gates) to optimize space usage or enhance the performance of user equipment. A processor can also be implemented as a combination of computing processing units. In the present disclosure, terms such as "repository", "storage device", "data repository", "data storage device", "database", and substantially any other information storage component related to the operation and functionality of a component are used to refer to a "memory component", an entity embodied in "memory", or a component that includes memory. It should be understood that the memory or memory component described herein can be volatile memory or non-volatile memory, or can include both volatile memory and non-volatile memory. By way of illustration and not limitation, non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM), flash memory, or non-volatile random access memory (RAM) (e.g., ferroelectric RAM (FeRAM)). For example, volatile memory can include RAM that can act as an external cache memory. By way of illustration and not limitation, RAM can be provided in various forms, such as synchronous RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), direct Rambus RAM (DRRAM), direct Rambus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM). Additionally, the disclosed memory components of the systems or computer-implemented methods herein are intended to include but not be limited to including these and any other suitable types of memory.

[0212] The foregoing description includes only examples of systems and computer-implemented methods. Of course, it is not possible to describe every conceivable combination of components or computer-implemented methods for the purposes of describing the present disclosure, but many other combinations and permutations are possible. Additionally, to the extent that the terms "including," "having," "owning," etc. are used in the detailed description, the claims, the appendices, and the drawings, such terms are intended to be inclusive in a manner similar to the term "comprising," as "comprising" is interpreted when used as a transitional word in the claims.

[0213] The descriptions of the various embodiments have been presented for purposes of illustration, but are not intended to be exhaustive or limited to the disclosed embodiments. Many modifications and variations will be apparent without departing from the scope and spirit of the described embodiments. The terms used herein were chosen to best explain the principles of the embodiments, the practical application or technical improvement over technologies found in the marketplace, or to enable other ordinary skilled artisans in the art to understand the embodiments disclosed herein.

Claims

1. A system, the system comprising: A processor (e.g., 108), the processor executing computer-executable components stored in a non-transitory computer-readable memory (e.g., 110), wherein the computer-executable components include: An access component (e.g., 112), the access component accessing a plurality of electronic documents (e.g., 106) associated with the design or manufacture of a medical imaging scanner (e.g., 104); and A knowledge graph component (e.g., 114), the knowledge graph component constructing a knowledge graph (e.g., 206) representing the plurality of electronic documents by iteratively executing a generative text-to-text neural network (e.g., 202) on a design discovery tree (e.g., 204) associated with the medical imaging scanner.

2. The system according to claim 1, wherein the access component accesses a natural language query (e.g., 1602) regarding the medical imaging scanner, and wherein the computer-executable components further include: A query component (e.g., 116), the query component converting the natural language query into a structured query (e.g., 1606) via the execution of another neural network (e.g., 1604), and executing the structured query on the knowledge graph, thereby generating an electronic answer (e.g., 1608) to the natural language query.

3. The system according to claim 1, wherein for each element of the design discovery tree, the element is a text prompt (e.g., 308) requesting an identification of corresponding technical design details (e.g., 310) of the medical imaging scanner.

4. The system according to claim 3, wherein the text prompt has a known relationship (e.g., 306) with another text prompt (e.g., 302) of the design discovery tree.

5. The system according to claim 4, wherein the knowledge graph component inserts a node shell (e.g., 802) into the knowledge graph, attaches an instantiation of the known relationship to the node shell, and fills the node shell by executing the generative text-to-text neural network on the text prompt in a retrieval-augmented generation manner.

6. The system according to claim 5, wherein the knowledge graph component executes the generative text-to-text neural network on the text prompt in a retrieval-augmented generation manner by: Identifying one or more first electronic documents (e.g., 902) coherent with the text prompt from the plurality of electronic documents; Cascading the text prompt with the one or more first electronic documents, thereby generating a cascade; and Executing the generative text-to-text neural network on the cascade, such that the generative text-to-text neural network synthesizes text content (e.g., 1002) describing the corresponding technical design details of the medical imaging scanner.

7. The system according to claim 6, wherein the knowledge graph component fills the node shell with the text content.

8. The system according to claim 1, wherein the plurality of electronic documents includes: A blueprint or schematic associated with the medical imaging scanner; or a failure analysis report associated with the medical imaging scanner.

9. A computer-implemented method, the computer-implemented method comprising: accessing, by a device (e.g., via 112) operatively coupled to a processor (e.g., 108), a plurality of electronic documents (e.g., 106) associated with the design or manufacture of a medical imaging scanner (e.g., 104); and constructing, by the device (e.g., via 114), a knowledge graph (e.g., 206) representing the plurality of electronic documents by iteratively performing a generative text-to-text neural network (e.g., 202) on a design discovery tree (e.g., 204) associated with the medical imaging scanner.

10. The computer-implemented method according to claim 9, the computer-implemented method further comprising: accessing, by the device (e.g., via 112), a natural language query (e.g., 1602) regarding the medical imaging scanner; converting, by the device (e.g., via 116), the natural language query into a structured query (e.g., 1606) via the execution of another neural network (e.g., 1604); and performing, by the device (e.g., via 116), the structured query on the knowledge graph, thereby generating an electronic answer (e.g., 1608) to the natural language query.

11. The computer-implemented method according to claim 9, wherein for each element of the design discovery tree, the element is a text prompt (e.g., 308) requesting an identification of corresponding technical design details (e.g., 310) of the medical imaging scanner.

12. The computer-implemented method according to claim 11, wherein the text prompt has a known relationship (e.g., 306) with another text prompt (e.g., 302) of the design discovery tree.

13. The computer-implemented method according to claim 12, the computer-implemented method further comprising: inserting, by the device (e.g., via 114), a node shell (e.g., 802) into the knowledge graph; attaching, by the device (e.g., via 114), an instantiation of the known relationship to the node shell; and and populating, by the device (e.g., via 114), the node shell by performing the generative text-to-text neural network on the text prompt in a retrieval-augmented generation manner.

14. The computer-implemented method according to claim 13, wherein performing the generative text-to-text neural network on the text prompt in a retrieval-augmented generation manner comprises: identifying, by the device (e.g., via 114), one or more first electronic documents (e.g., 902) from the plurality of electronic documents that are relevant to the text prompt; cascading, by the device (e.g., via 114), the text prompt with the one or more first electronic documents, thereby generating a cascade; and performing, by the device (e.g., via 114), the generative text-to-text neural network on the cascade, thereby causing the generative text-to-text neural network to synthesize text content (e.g., 1002) describing the corresponding technical design details of the medical imaging scanner.

15. The computer-implemented method according to claim 14, wherein the device fills the node housing with the text content.

Images