Explainable and automated decisions in computer-based reasoning systems

• US 10,528,877 B1
• Filed: 11/30/2018
• Issued: 01/07/2020
• Est. Priority Date: 11/13/2018
• Status: Active Grant

• Current Assignee: Howso Incorporated
• Original Assignee: Diveplane Corporation
• Inventors: Hazard, Christopher James, Fusting, Christopher, Resnick, Michael
• Primary Examiner(s): Misir, Dave
• Application Number:
  US16/205,373
• Time in Patent Office: 403 Days
• Field of Search: 706 46, 706 47
• CPC Class Codes: G05D 1/0061 for transition from automatic pilot to manual pilot and vice versa
• CPC Class Codes: G05D 1/0088 characterized by the autonomous decision making process, e.g. artificial intelligence, predefined behaviours using knowledge based models G06N5/00
• CPC Class Codes: G05D 1/0221 involving a learning process
• CPC Class Codes: G06F 18/2415 based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/04 Inference or reasoning models
• CPC Class Codes: G06N 5/045 Explanation of inference; Explainable artificial intelligence [XAI]; Interpretable artificial intelligence
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

The techniques herein include using an input context to determine a suggested action. One or more explanations may also be determined and returned along with the suggested action. The one or more explanations may include (i) one or more most similar cases to the suggested case (e.g., the case associated with the suggested action) and, optionally, a conviction score for each nearby cases; (ii) action probabilities, (iii) excluding cases and distances, (iv) archetype and/or counterfactual cases for the suggested action; (v) feature residuals; (vi) regional model complexity; (vii) fractional dimensionality; (viii) prediction conviction; (ix) feature prediction contribution; and/or other measures such as the ones discussed herein, including certainty. In some embodiments, the explanation data may be used to determine whether to perform a suggested action.

Claims

1. A method comprising:

receiving a request for a suggested action based on an input context, in a case-based reasoning system, wherein the case-based reasoning system includes a case-based reasoning model;

determining one or more candidate cases based on the input context in the case-based reasoning system, wherein the one or more candidate cases include respective one or more candidate actions;

determining the suggested action based on the respective one or more candidate actions;

determining a certainty score based on the suggested action, wherein the certainty score is determined based on a conviction function associated with;

removing a suggested case associated with the suggested action from the case-based reasoning model; and

adding the suggested case into the case-based reasoning model,

wherein the conviction function is a measure of how much information the suggested case distorts the case-based reasoning model;

responding to the request for suggested action with the suggested action and the certainty score;

when the certainty score is beyond a certain threshold, causing control of a controllable system based on the suggested action;

when the certainty score is not beyond the certain threshold;

determining one or more explanation factors for the suggested action determined based at least in part on the input context;

providing the one or more explanation factors in response to the request for the suggested action;

wherein the method is performed by one or more computing devices.

2. The method of claim 1, wherein the suggested action is a suggested labeling, and the input context is a current context for an image labeling system during operation of the image labeling system, and causing performance of the suggested action comprises causing the suggested labeling to be performed by the image labeling system.

3. The method of claim 1, wherein the input context is a current context for a self-driving car during operation of the self-driving car, and causing performance of the suggested action comprises causing the suggested action to be performed by the self-driving car.

4. The method of claim 1, further comprising:

when the certainty score is not beyond the certain threshold, after providing the suggested action and the one or more explanation factors in response to the request for the suggested action, receiving an indication whether to perform the suggested action;

when the received indication indicates performance of the suggested action, causing control of the controllable system based on the suggested action.

5. The method of claim 1, wherein determining the suggested action comprises:

determining a weighting for each action of the respective one or more candidate actions based on a function of a distance between the input context and each of the one or more candidate cases;

determining the suggested action based on the weighting for each action of the respective one or more candidate actions.

6. The method of claim 1, wherein determining the certainty score comprises determining a prediction conviction value for the suggested case.

7. The method of claim 1, wherein the suggested action is associated with a suggested case and determining the certainty score for the suggested action comprises:

determining information required to describe a position of the suggested case relative to cases in the case-based reasoning model.

8. A system for performing a machine-executed operation involving instructions, wherein said instructions are instructions which, when executed by one or more computing devices, cause performance of a process comprising:

receiving a request for a suggested action based on an input context, in a case-based reasoning system, wherein the case-based reasoning system includes a case-based reasoning model;

determining one or more candidate cases based on the input context in the case-based reasoning system, wherein the one or more candidate cases include respective one or more candidate actions;

determining the suggested action based on the respective one or more candidate actions;

determining a certainty score based on the suggested action, wherein the certainty score is determined based on a conviction function associated with;

removing a suggested case associated with the suggested action from the case-based reasoning model; and

adding the suggested case into the case-based reasoning model,

wherein the conviction function is a measure of how much information is required to describe a position of the suggested case relative to existing cases in the case-based reasoning model;

responding to the request for suggested action with the suggested action and the certainty score;

when the certainty score is beyond a certain threshold, causing control of a controllable system based on the suggested action;

when the certainty score is not beyond the certain threshold;

determining one or more explanation factors for the suggested action determined based at least in part on the input context;

providing the one or more explanation factors in response to the request for the suggested action.

9. The system of claim 8, wherein the suggested action is a suggested labeling, and the input context is a current context for an image labeling system during operation of the image labeling system, and causing performance of the suggested action comprises causing the suggested labeling to be performed by the image labeling system.

10. The system of claim 8, wherein the input context is a current context for a self-driving car during operation of the self-driving car, and causing performance of the suggested action comprises causing the suggested action to be performed by the self-driving car.

11. The system of claim 8, further comprising:

when the certainty score is not beyond the certain threshold, after providing the suggested action and the one or more explanation factors in response to the request for the suggested action, receiving an indication whether to perform the suggested action;

when the received indication indicates performance of the suggested action, causing control of the controllable system based on the suggested action.

12. The system of claim 8, wherein determining the suggested action comprises:

determining a weighting for each action of the respective one or more candidate actions based on a function of a distance between the input context and each of the one or more candidate cases;

determining the suggested action based on the weighting for each action of the respective one or more candidate actions.

13. The system of claim 8, wherein determining the certainty score comprises determining a prediction conviction value for the suggested case.

14. The system of claim 8, wherein the suggested action is associated with a suggested case and determining the certainty score for the suggested action comprises:

determining information required to describe a position of the suggested case relative to cases in the case-based reasoning model.

15. A method comprising:

receiving a request for a suggested action based on an input context, in a case-based reasoning system, wherein the case-based reasoning system includes a case-based reasoning model;

determining one or more candidate cases based on the input context in the case-based reasoning system, wherein the one or more candidate cases include respective one or more candidate actions;

determining the suggested action based on the respective one or more candidate actions,

wherein the suggested action is associated with a suggested case, wherein the suggested case has multiple features;

determining a certainty score based on the suggested action, wherein the certainty score is determined based on a conviction function associated with;

determining feature prediction contributions of the multiple features of the suggested case;

holding each feature of the multiple features out from the case-based reasoning model; and

including each feature of the multiple features in the case-based reasoning model;

wherein the conviction function is a measure of how important the multiple features of the suggested case are in making predictions in the case-based reasoning model;

responding to the request for suggested action with the suggested action and the feature prediction contributions;

when the feature prediction contributions for certain features of the multiple features are beyond a certain threshold, causing control of a controllable system based on the suggested action;

when the feature prediction contributions for the certain features are not beyond the certain threshold;

determining one or more explanation factors for the suggested action determined based at least in part on the input context;

providing the one or more explanation factors in response to the request for the suggested action;

wherein the method is performed by one or more computing devices.

16. The method of claim 15, wherein the certain features are features in an undesirable feature prediction list and the method further comprises:

determining whether the feature prediction contributions for undesirable features in the undesirable feature prediction list are not below the certain threshold;

when the feature prediction contributions for the undesirable features are not below the certain threshold;

determining the one or more explanation factors for the suggested action determined based at least in part on the feature prediction contributions;

providing the one or more explanation factors in response to the request for the suggested action.

17. The method of claim 15, wherein the certain features are features in a desirable feature prediction list and the method further comprises:

determining whether the feature prediction contributions for desirable features in the desirable feature prediction list are not above the certain threshold;

when the feature prediction contributions for the desirable features are not above the certain threshold;

determining the one or more explanation factors for the suggested action determined based at least in part on the feature prediction contributions;

providing the one or more explanation factors in response to the request for the suggested action.

18. The method of claim 15, wherein the suggested action is a suggested labeling, and the input context is a current context for an image labeling system during operation of the image labeling system, and causing performance of the suggested action comprises causing the suggested labeling to be performed by the image labeling system.