# US 2020/0394529 A1 · Knowledge engineering and reasoning on a knowledge graph

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2020/0394529 A1
- Document ID: US20200394529A1
- Status: Published application
- Kind: A1
- Application number: 16/900,369
- Filed: 2020-06-12
- Published: 2020-12-17
- Earliest claimed: 2019-06-14
- Assignees: Accenture Global Solutions Limited
- Applicants: Accenture Global Solutions Limited
- Inventors: Zhijie Wang; William Richard Gatehouse; Teresa Sheausan Tung
- Web page: https://patentgrep.com/patent/US20200394529A1

## CPC

- G06N 5/022: Knowledge engineering; Knowledge acquisition
- G06F 18/22: Matching criteria, e.g. proximity measures
- G06F 18/2415: based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
- G06F 18/2431: Multiple classes
- G06F 18/29: {Graphical models, e.g. Bayesian networks}
- G06N 5/02: Knowledge representation; Symbolic representation
- G06N 5/046: Forward inferencing; Production systems
- G06N 20/00: Machine learning
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A model management tool is provided for performing analysis on data in a knowledge graph representation and enforcing data standardization to increase performance when reusing existing models to develop new artificial intelligence applications.

## Claims

### Claim 1

A method comprising:
at a capture layer of a model repository stack:
accessing, via network interface circuitry, a project knowledge graph, the project knowledge graph including nodes representing multiple tasks, the multiple tasks determined via deconstruction of an end-to-end artificial intelligence (AI) application into the multiple tasks; and
for an individual one of the multiple tasks, at classification circuitry:
traversing the project knowledge graph to determine a corresponding AI task block including contextualized execution code, the contextualized execution code configured to perform the individual one of the multiple tasks in a specific context defined for the AI application on the project knowledge graph;
capturing a portion of the specific context from the project knowledge graph; and
after a determination that capture of the specific context was incomplete, re-traversing the project knowledge graph to capture an additional portion of the specific context;
at a repository graph layer of the model repository stack:
for the corresponding AI task block:
storing the corresponding AI task block in a repository;
importing the portion of the specific context to a repository knowledge graph by appending a portion of the project knowledge graph to the repository knowledge graph;
applying a first descriptor rule to the to the portion of the specific context to determine that the capture of the specific context was incomplete; and
after the capture is completed, expanding the repository knowledge graph to include the additional potion of the specific context;
associating the corresponding AI task block with an AI classification based on the individual one of the multiple tasks and the specific context;
at a deployment layer of the model repository stack;
based on the AI classification, the specific context, the individual one of the multiple tasks, or a combination thereof, returning the corresponding AI task block as an option in response to a query; and
sending the corresponding AI task block to a requester responsive to a selection of the option.

### Claim 2 (depends on 1)

The method of claim 1 further including expanding the AI classification to cover the individual one of the multiple tasks, the specific context, or both.

### Claim 3 (depends on 2)

The method of claim 2 where expanding the AI classification includes super-classing an existing class by defining a super-class covering the existing class as a subclass and the corresponding AI task block, the specific context, or both outside the subclass.

### Claim 4 (depends on 3)

The method of claim 3, where super-classing further includes causing reverse-inheritance of parameters from the subclass to the super-class upon creation of the super-class.

### Claim 5 (depends on 4)

The method of claim 4, further causing forward-inheritance of parameters from the super-class to the subclass for parameters applied after creation of the super-class.

### Claim 6 (depends on 3)

The method of claim 3, further including determining not to reverse-inherit a selected parameter from the subclass responsive to the specific context, the corresponding AI task block, or both.

### Claim 7 (depends on 1)

The method of claim 1, where applying the first descriptor rule includes traversing the specific context across a transformation from an input type to an output type.

### Claim 8 (depends on 7)

The method of claim 7, where the determination that the capture of the specific context was incomplete includes determining that a node for the transformation, a node for the input type, a node for the output type, or any combination thereof is missing from the portion of the specific context.

### Claim 9 (depends on 8)

The method of claim 8, further comprising identifying a node missing from the portion based on a node present in the portion that is part of an incomplete input-transformation-output tuple.

### Claim 10 (depends on 1)

The method of claim 1, before returning the corresponding AI task block as an option in response to the query, determining to respond to the query with an AI task block associated with the AI classification.

### Claim 11 (depends on 10)

The method of claim 10, before returning the corresponding AI task block as an option in response to the query:
evaluating AI task blocks associated with the AI classification including the corresponding AI task block; and
determining a structural similarity between the corresponding AI task block and a term from the query.

### Claim 12

A system comprising:
network interface circuitry configured to:
at a capture layer of a model repository stack:
accessing, via network interface circuitry, a project knowledge graph, the project knowledge graph including nodes representing multiple tasks, the multiple tasks determined via deconstruction of an end-to-end artificial intelligence (AI) application into the multiple tasks; and
classification circuitry in data communication with the network interface circuitry, the classification circuitry configured to:
at the capture layer:
for an individual one of the multiple tasks:
traversing the project knowledge graph to determine a corresponding AI task block including contextualized execution code, the contextualized execution code configured to perform the individual one of the multiple tasks in a specific context defined for the AI application on the project knowledge graph;
capturing a portion of the specific context from the project knowledge graph; and
after a determination that capture of the specific context was incomplete, re-traversing the project knowledge graph to capture an additional portion of the specific context;
at a repository graph layer of the model repository stack:
for the corresponding AI task block:
storing the corresponding AI task block in a repository;
importing the portion of the specific context to a repository knowledge graph by appending a portion of the project knowledge graph to the repository knowledge graph;
applying a first descriptor rule to the to the portion of the specific context to determine that the capture of the specific context was incomplete; and
after the capture is completed, expanding the repository knowledge graph to include the additional potion of the specific context;
associating the corresponding AI task block with an AI classification based on the individual one of the multiple tasks and the specific context;
deployment circuitry configured to:
at a deployment layer of the model repository stack;
based on the AI classification, the specific context, the individual one of the multiple tasks, or a combination thereof, returning the corresponding AI task block as an option in response to a query; and
sending, via the network interface circuitry, the corresponding AI task block to a requester responsive to a selection of the option.

### Claim 13 (depends on 12)

The system of claim 12, where the classification circuitry is further configured to expand the AI classification to cover the individual one of the multiple tasks, the specific context, or both.

### Claim 14 (depends on 13)

The system of claim 13 where the classification circuitry is further configured to expand the AI classification by super-classing an existing class by defining a super-class covering the existing class as a subclass and the corresponding AI task block, the specific context, or both outside the subclass.

### Claim 15 (depends on 14)

The system of claim 14, where the classification circuitry is further configured to super-class by causing reverse-inheritance of parameters from the subclass to the super-class upon creation of the super-class.

### Claim 16 (depends on 15)

The system of claim 15, where the classification circuitry is further configured to cause forward-inheritance of parameters from the super-class to the subclass for parameters applied after creation of the super-class.

### Claim 17 (depends on 14)

The system of claim 14, further including determining not to revere-inherit a selected parameter from the subclass responsive to the specific context, the corresponding AI task block, or both.

### Claim 18

A product comprising:
machine-readable media other than a transitory signal; and
instructions stored on the machine-readable media, the instructions configured to cause a machine to:
at a capture layer of a model repository stack:
access, via network interface circuitry, a project knowledge graph, the project knowledge graph including nodes representing multiple tasks, the multiple tasks determined via deconstruction of an end-to-end artificial intelligence (AI) application into the multiple tasks; and
for an individual one of the multiple tasks, at classification circuitry:
traverse the project knowledge graph to determine a corresponding AI task block including contextualized execution code, the contextualized execution code configured to perform the individual one of the multiple tasks in a specific context defined for the AI application on the project knowledge graph;
capture a portion of the specific context from the project knowledge graph; and
after a determination that capture of the specific context was incomplete, re-traverse the project knowledge graph to capture an additional portion of the specific context;
at a repository graph layer of the model repository stack:
for the corresponding AI task block:
store the corresponding AI task block in a repository;
import the portion of the specific context to a repository knowledge graph by appending a portion of the project knowledge graph to the repository knowledge graph;
apply a first descriptor rule to the to the portion of the specific context to determine that the capture of the specific context was incomplete; and
after the capture is completed, expand the repository knowledge graph to include the additional potion of the specific context;
associate the corresponding AI task block with an AI classification based on the individual one of the multiple tasks and the specific context;
at a deployment layer of the model repository stack;
based on the AI classification, the specific context, the individual one of the multiple tasks, or a combination thereof, return the corresponding AI task block as an option in response to a query; and
send the corresponding AI task block to a requester responsive to a selection of the option.

### Claim 19 (depends on 18)

The product of claim 18, where the instructions are further configured to cause the machine to: before returning the corresponding AI task block as an option in response to the query, determine to respond to the query with an AI task block associated with the AI classification.

### Claim 20 (depends on 19)

The product of claim 19, where the instructions are further configured to cause the machine to:
before returning the corresponding AI task block as an option in response to the query:
evaluate AI task blocks associated with the AI classification including the corresponding AI task block; and
determine a structural similarity between the corresponding AI task block and a term from the query.
