Classifier Learning with Causal Effect Variance Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for learning classifiers to ensure fair decision-making by removing causal effects between sensitive features and decision results often result in large individual causal effects, leading to non-fair decisions despite a zero mean value for the group.
Innovation Solution
A learning apparatus that inputs training data and a causal graph, solving a constrained optimization problem to ensure the mean of causal effects is within a predetermined range and the variance is equal to or smaller than a specified value, using a weakly convex objective function to guarantee convergence and remove causal effects in each training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a constraint is imposed such that the mean value of causal effects is zero for the entire group, then the overall fairness is improved, but the variance of causal effects becomes large causing some individuals to receive discriminatory decisions
Solution Approach 1:
The patent changes the optimization parameters from only constraining the mean causal effect to constraining both the mean and variance of causal effects. By adding the variance constraint parameter, the system achieves better individual-level fairness while maintaining overall fairness, resolving the contradiction between group-level and individual-level fairness requirements
Solution Approach 2:
The patent applies prior cushioning by pre-constraining the variance of causal effects during the classifier learning process. This prevents large individual causal effects from occurring in the first place, rather than allowing them to occur and then correcting them later, thus ensuring individual fairness is maintained throughout the decision-making process
Data Source
AI summary
A learning apparatus according to an embodiment includes: input means for inputting training data for learning a classifier and a causal graph representing causal relationships between variables included in the training data; and learning means for learning the classifier by solving a constrained optimization problem in which a mean of causal effects between predetermined variables is within a predetermined range and a variance of the causal effects is equal to or smaller than a predetermined value, using the training data and the causal graph input by the input means.