DISCO: Mitigating Bias in Deep Learning with Conditional Distance Correlation

ArXi:2506.11653v3 Announce Type: replace-cross Dataset bias often leads deep learning models to exploit spurious correlations instead of task-relevant signals. We