IP Library Granted Patent US 11,593,680
Granted Patent B2
US 11,593,680 · App. 16/928,171 · Granted Feb 28, 2023

Predictive models having decomposable hierarchical layers configured to generate interpretable results

Inventors: Nianjun Zhou (Yorktown Heights, NY); Wesley M. Gifford (Ridgefield, CT); Ta-Hsin Li (Danbury, CT); Pietro Mazzoleni (New York, NY); Pavankumar Murali (Ardsley, NY)
Assignee: International Business Machines Corporation
G06N5/04G06N20/00
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 11,593,680
App. No.
16/928,171
Granted
Feb 28, 2023
Kind
B2
Abstract

A computer-implemented method for providing interpretable predictions from a machine learning model includes receiving a data structure that represents a hierarchical structure of a set of features (X) used by one or more predictive models to generate a set of predictions (Y). An interpretability model is built corresponding to the predictive models, by assigning an interpretability to each prediction Y i based on the hierarchical structure. Assigning the interpretability includes decomposing X into a plurality of partitions X j using the hierarchical structure, wherein X=U 1 N X j , N being the number of partitions. Further, each partition is decomposed into a plurality of sub-partitions using the hierarchical structure until atomic sub-partitions are obtained. A score is computed for each partition as a function of the predicted scores of the sub-partitions, wherein the predicted scores represent interactions between the sub-partitions. Further, an interpretation of a prediction is outputted.

Claims (62)

1. A computer-implemented method for providing interpretable predictions from a machine learning model for determining whether a patient qualifies for a medial trial for a particular diagnosis, the method comprising:

receiving, by a processor of a computing device, a data structure that represents a hierarchical structure of a set of features (X) of that is used by one or more predictive models to generate a set of predictions (Y); and

building, by the processor, an interpretability model corresponding to the predictive models, by assigning an interpretability to each prediction Y i of the predictive model based on the hierarchical structure, wherein assigning the interpretability comprises:

decomposing, by the processor, the set of features (X) of the input data into a plurality of partitions X j using the hierarchical structure, wherein X=U 1 N X j , N being the number of partitions;

decomposing, by the processor, each partition from the partitions into a plurality of sub-partitions using the hierarchical structure until atomic sub-partitions are obtained; and

computing, by the processor, a score for each partition as a function of the predicted scores of the sub-partitions of said each partition, wherein the predicted scores represent interactions between the sub-partitions; and

outputting, by the processor, an interpretation of a prediction provided by the predictive model based on the interpretability model, wherein the interpretation represents a weightage of the one or more features used by the predictive model to generate the prediction,

wherein:

the method further comprises identifying, by the processor, a recommended action based on the predicted score for a selected feature,

the predicted score for the selected feature indicates that a patient qualifies for a medical trial for a particular diagnosis, and

the recommended action comprises:

presenting a notification to a user that the patient qualifies for the medial trial; and

upon a selection to proceed by the user, generating an email outlining a specific type of medical treatment available through the medical trial.

2. The computer-implemented method of claim 1 , wherein the partitions of features are mutually exclusive subsets of the input data.

3. The computer-implemented method of claim 1 further comprising generating an overall score for the input data based on the predicted score for each partition.

4. The computer-implemented method of claim 1 , wherein generating the hierarchical structure further comprises:

receiving, by the processor, industry data that is curated by one or more experts; and

generating, by the processor, the hierarchical structure using the industry data by constructing a tree model by parsing the industry data into the partitions of features and creating associations between the partitions.

5. The computer-implemented method of claim 1 , wherein each partition is associated with a corresponding interpretability model.

6. The computer-implemented method of claim 1 , wherein the interpretability model is an attention model.

7. A system comprising:

at least one memory storing computer-executable instructions; and

at least one processor configured to access the at least one memory and perform a method for providing interpretable predictions from a machine learning model for determining whether a patient qualifies for a medial trial for a particular diagnosis, the method comprising:

receiving, by a processor of a computing device, a data structure that represents a hierarchical structure of a set of features (X) used by one or more predictive models to generate a set of predictions (Y); and

building, by the processor, an interpretability model corresponding to the predictive models, by assigning an interpretability to each prediction Y i of the predictive model based on the hierarchical structure, wherein assigning the interpretability comprises:

decomposing, by the processor, the set of features (X) of the input data into a plurality of partitions X j using the hierarchical structure, wherein X=U 1 N X j , N being the number of partitions;

decomposing, by the processor, each partition from the partitions into a plurality of sub-partitions using the hierarchical structure until atomic sub-partitions are obtained; and

computing, by the processor, a score for each partition as a function of the predicted scores of the sub-partitions of said each partition, wherein the predicted scores represent interactions between the sub-partitions; and

outputting, by the processor, an interpretation of a prediction provided by the predictive model based on the interpretability model, wherein the interpretation represents a weightage of the one or more features used by the predictive model to generate the prediction,

wherein:

the method further comprises identifying, by the processor, a recommended action based on the predicted score for a selected feature,

the predicted score for the selected feature indicates that a patient qualifies for a medical trial for a particular diagnosis, and

the recommended action comprises:

presenting a notification to a user that the patient qualifies for the medial trial; and

upon a selection to proceed by the user, generating an email outlining a specific type of medical treatment available through the medical trial.

8. The system of claim 7 , wherein the partitions of features are mutually exclusive subsets of the input data.

9. The system of claim 7 , wherein the method further comprises generating an overall score for the input data based on the predicted score for each partition.

10. The system of claim 7 , wherein generating the hierarchical structure further comprises:

receiving industry data that is curated by one or more experts; and

generating the hierarchical structure using the industry data by constructing a tree model by parsing the industry data into the partitions of features and creating associations between the partitions.

11. The system of claim 7 , wherein each partition is associated with a corresponding interpretability model.

12. The system of claim 7 , wherein the interpretability model is an attention model.

13. A computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by one or more processors to cause the one or more processors to perform a method for providing interpretable predictions from a machine learning model for determining whether a patient qualifies for a medial trial for a particular diagnosis comprising:

receiving, by a processor of a computing device, a data structure that represents a hierarchical structure of a set of features (X) used by one or more predictive models to generate a set of predictions (Y);

building, by the processor, an interpretability model corresponding to the predictive models, by assigning an interpretability to each prediction Y i of the predictive model based on the hierarchical structure, wherein assigning the interpretability comprises:

decomposing, by the processor, the set of features (X) of the input data into a plurality of partitions X j , using the hierarchical structure, wherein X=U 1 N X j , N being the number of partitions;

decomposing, by the processor, each partition from the partitions into a plurality of sub-partitions using the hierarchical structure until atomic sub-partitions are obtained; and

computing, by the processor, a score for each partition as a function of the predicted scores of the sub-partitions of said each partition, wherein the predicted scores represent interactions between the sub-partitions; and

outputting, by the processor, an interpretation of a prediction provided by the predictive model based on the interpretability model, wherein the interpretation represents a weightage of the one or more features used by the predictive model to generate the prediction,

wherein:

the method further comprises identifying, by the processor, a recommended action based on the predicted score for a selected feature,

the predicted score for the selected feature indicates that a patient qualifies for a medical trial for a particular diagnosis, and

the recommended action comprises:

presenting a notification to a user that the patient qualifies for the medial trial; and

upon a selection to proceed by the user, generating an email outlining a specific type of medical treatment available through the medical trial.

14. The computer program product of claim 13 , wherein the partitions of features are mutually exclusive subsets of the input data.

15. The computer program product of claim 13 , wherein the method further comprises generating an overall score for the input data based on the predicted score for each partition.

16. The computer program product of claim 13 , wherein generating the hierarchical structure further comprises:

receiving industry data that is curated by one or more experts; and

generating the hierarchical structure using the industry data by constructing a tree model by parsing the industry data into the partitions of features and creating associations between the partitions.

17. The computer program product of claim 13 , wherein each partition is associated with a corresponding interpretability model.

18. The computer program product of claim 13 , wherein the interpretability model is an attention model.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Jul 14, 2020
From: ZHOU, NIANJUN; GIFFORD, WESLEY M.; LI, TA-HSIN; MAZZOLENI, PIETRO; MURALI, PAVANKUMAR
To: INTERNATIONAL BUSINESS MACHINES CORPORATION
Reel/Frame 053203/0941 →
Continuity (1)
Related Publication 20220019911A1 · Jan 20, 2022
Cited By (1)
US 12,423,619