DISCO: Mitigating Bias in Deep Learning with Conditional Distance Correlation

2025-06-13

AI Summary

Deep learning models often rely on spurious, causally irrelevant features—such as lighting conditions in images—for prediction, leading to biased decisions and poor out-of-distribution generalization. To address this, we propose a *counterfactual prediction* framework that enforces classifiers to utilize only the direct causal path from label to image. Our method introduces two key innovations: (1) the first application of conditional independence optimization to bias mitigation, realized via a novel DISCO regularizer based on Conditional Distance Correlation (CDC); and (2) the construction of a Standard Anticausal Model (SAM), which theoretically guarantees counterfactual invariance. Evaluated across multiple regression and classification benchmarks, our approach matches or surpasses classical kernel-based methods in accuracy, while offering superior scalability, theoretical interpretability, and empirical robustness.

Technology Category

• Machine Learning: Causal Learning
• Computer Vision: Bias, Fairness & Privacy
• Reasoning under Uncertainty: Causality

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Responsible Web: Algorithmic accountability and transparency on the web

Abstract

During prediction tasks, models can use any signal they receive to come up with the final answer - including signals that are causally irrelevant. When predicting objects from images, for example, the lighting conditions could be correlated to different targets through selection bias, and an oblivious model might use these signals as shortcuts to discern between various objects. A predictor that uses lighting conditions instead of real object-specific details is obviously undesirable. To address this challenge, we introduce a standard anti-causal prediction model (SAM) that creates a causal framework for analyzing the information pathways influencing our predictor in anti-causal settings. We demonstrate that a classifier satisfying a specific conditional independence criterion will focus solely on the direct causal path from label to image, being counterfactually invariant to the remaining variables. Finally, we propose DISCO, a novel regularization strategy that uses conditional distance correlation to optimize for conditional independence in regression tasks. We can show that DISCO achieves competitive results in different bias mitigation experiments, deeming it a valid alternative to classical kernel-based methods.

Problem

• Mitigating bias in deep learning models
• Eliminating causally irrelevant signals in predictions
• Ensuring conditional independence in regression tasks

Innovation

• SAM model creates causal framework for bias analysis
• Classifier focuses on direct causal path from label
• DISCO uses conditional distance correlation for independence