Realistic counterfactual explanation of machine learning predictions

ChangeBridge: Patent Grants - AI & Computing (G06N)

Published April 7th, 2026

Detected April 7th, 2026

Summary

USPTO granted IBM patent US12596941B2 for a computer-implemented method of generating counterfactual explanations for machine learning model predictions using Constraint Satisfaction Problem (CSP) solvers. The method analyzes ML datasets to deduce constraints, generates CSP rules, and uses solver solutions as counterfactual explanations for model predictions. The patent contains 19 claims and covers technology for improving ML model interpretability.

View original document View source feed page

What changed

USPTO issued patent grant US12596941B2 to International Business Machines Corporation for a method of explaining machine learning predictions through counterfactual analysis. The patented technology automatically analyzes ML datasets to deduce feature constraints, generates two sets of CSP rules (one for dataset constraints, one for perturbation candidates), and solves the formulated CSP to produce counterfactual explanations.

For technology companies and AI developers, this patent establishes IBM's proprietary method for ML explainability using constraint satisfaction problem solvers. Organizations developing similar counterfactual explanation systems should review potential patent infringement implications and consider licensing discussions with IBM.

What to do next

Monitor for updates

Source document (simplified)

← USPTO Patent Grants

Realistic counterfactual explanation of machine learning predictions

Grant US12596941B2 Kind: B2 Apr 07, 2026

Assignee

INTERNATIONAL BUSINESS MACHINES CORPORATION

Inventors

Michael Vinov, Oleg Blinder, Diptikalyan Saha, Sandeep Hans, Aniya Aggarwal, Omer Yehuda Boehm, Eyal Bin

Abstract

A computer-implemented method comprising, automatically: analyzing a machine learning dataset which comprises multiple datapoints, to deduce constraints on features of the datapoints; generating a first set of CSP (Constraint Satisfaction Problem) rules expressing the constraints; based on a machine learning model which was trained on the dataset, generating a second set of CSP rules that define one or more perturbation candidates among the features of one of the datapoints; formulating a CSP based on the first and second sets of CSP rules; solving the formulated CSP using a solver; and using the solution of the CSP as a counterfactual explanation of a prediction made by the machine learning model with respect to the one datapoint.