AI governance using tamper proof model metrics
AssigneeInternational Business Machines Corporation
InventorsManish Anand Bhide, Ravi Chandra Chamarthy, Arunkumar Kalpathi Suryanarayanan
Life of this patent
Estimating the term…
Abstract
One example of a method comprises identifying a model to be validated that is stored in a repository; automatically computing and recording one or more model metrics for the model to be validated in a tamper-proof manner; comparing the computed tamper-proof metrics with one or more encoded rules and policies to determine if the model to be validated complies with the one or more encoded rules and policies; and outputting a notification to a device indicating a validation status of the model to be validated based on the comparison of the computed tamper-proof metrics with the one or more encoded rules and policies.
Claims19
A method comprising:
- an identifying operation selected from the group consisting of:
- a) identifying a model to be validated that is stored in a repository responsive to detecting that the model was pushed to the repository, and
- b) identifying the model to be validated comprises:
- detecting a change in the one or more encoded rules and policies;
- and
- in response to detecting the change, identifying the model to be validated based on the model having not been validated subsequent to the detected change;
- automatically computing and recording one or more model metrics for the model to be validated in a tamper-proof manner;
- comparing the computed tamper-proof metrics with one or more encoded rules and policies to determine if the model to be validated complies with the one or more encoded rules and policies;
- and
- outputting a notification to a device indicating a validation status of the model to be validated based on the comparison of the computed tamper-proof metrics with the one or more encoded rules and policies.
A method comprising:
- identifying a model to be validated that is stored in a repository;
- automatically computing and recording one or more model metrics for the model to be validated in a tamper-proof manner;
- comparing the computed tamper-proof metrics with one or more encoded rules and policies to determine if the model to be validated complies with the one or more encoded rules and policies;
- and
- outputting a notification to a device indicating a validation status of the model to be validated based on the comparison of the computed tamper-proof metrics with the one or more encoded rules and policies;
- wherein automatically computing and recording one or more model metrics for the model in a tamper-proof manner comprises:
- registering with the repository a service identification (ID) of a computation tool authorized to edit metadata of the model to be validated;
- and
- prohibiting edits to the metadata of the model to be validated that are not performed by the computation tool having the registered service ID.
A method comprising:
- identifying a model to be validated that is stored in a repository;
- automatically computing and recording one or more model metrics for the model to be validated in a tamper-proof manner;
- comparing the computed tamper-proof metrics with one or more encoded rules and policies to determine if the model to be validated complies with the one or more encoded rules and policies;
- and
- outputting a notification to a device indicating a validation status of the model to be validated based on the comparison of the computed tamper-proof metrics with the one or more encoded rules and policies;
- wherein comparing the computed tamper-proof metrics with one or more encoded rules and policies to determine if the model complies with the one or more encoded rules and policies includes:
- identifying training data specified in a label of the model to be validated that indicates the model to be validated was built with the training data;
- retrieving the identified training data;
- automatically building a sample model using the retrieved training data;
- automatically computing one or more metrics for the sample model;
- comparing the one or more metrics for the sample model with the one or more for the model to be validated;
- and
- verifying whether the model to be validated was built using the identified training data based on the comparison.
Dependencies are inferred from the claim text.