DISCO: Mitigating Bias in Deep Learning with Conditional Distance Correlation

Dataset bias often leads deep learning models to exploit spurious correlations instead of task-relevant signals. We introduce the Standard Anti-Causal Model (SAM), a unifying causal framework that characterizes bias mechanisms and yields a conditional independence criterion for causal stability. Building on this theory, we propose DISCO$_m$ and sDISCO, efficient and scalable estimators of conditional distance correlation that enable independence regularization in gradient-based models. Across six diverse datasets, our methods consistently outperform or are competitive in existing observed bias mitigation approaches, while requiring fewer hyperparameters and scaling seamlessly to multi-bias scenarios. This work bridges causal theory and practical deep learning, providing both a principled foundation and effective tools for robust prediction. Source Code: https://github.com/yakamoz5/DISCO.

Artificial intelligence models often cheat by learning unintended "shortcuts" from their training data instead of focusing on the actual task. For example, a model predicting a disease might wrongly rely on a patient's age rather than actual medical markers if age happens to be strongly tied to the illness in the dataset. To fix this, we created a structural framework to map out exactly how these biases trick AI models. Based on this understanding, we developed two highly efficient computational tools, DISCO_${m}$ and sDISCO. These tools act as a mathematical penalty system during training, forcing the AI to actively ignore known shortcuts and instead learn the true, robust patterns necessary to solve the problem. When tested across six diverse datasets, our methods consistently matched or outperformed existing bias-correction techniques. Importantly, they require fewer manual adjustments from developers and can seamlessly handle scenarios involving multiple overlapping biases. Ultimately, this research provides a practical toolkit for building fairer, more reliable AI systems that make decisions for the right reasons.