Systems and methods for counterfactual explanations in machine learning models
The two-stage method for counterfactual feature selection and optimization in machine learning models addresses the complexity of non-differentiable models, enhancing explainability and reliability of AI systems by providing detailed and efficient counterfactual explanations.
JP7597417B2Active Publication Date: 2024-12-10SALESFORCE INC
Patent Information
- Application Number
- JP2023521466
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2021-01-29
- Filing Date
- 2021-10-07
- Publication Date
- 2024-12-10
- Estimated Expiration
- 2041-10-07
Smart Images
Figure 0007597417000024 
Figure 0007597417000025 
Figure 0007597417000026
Abstract
The embodiments described herein provide a two-stage, model-independent approach to generating counterfactual explanations via counterfactual feature selection and counterfactual feature optimization. Given a query instance, counterfactual feature selection selects a subset of feature columns and values that could potentially change the prediction, and then counterfactual feature optimization determines the best feature values for the selected features as the counterfactual.
Need to check novelty before this filing date? Find Prior Art
Citation Information
Patent Citations
Clustering, Explainability, and Automated Decisions in Computer-Based Reasoning Systems
US20200134484A1
Information processing method, information processing device, and information processing computer program
WO2018225227A1