Modern machine learning models are complicated.

Existing bias mitigation methods for DNN models primarily work on learning debi-ased encoders.

"shortcut learning": when a predictor relies on shortcuts, i.e., spurious correlations between the inputs

"shortcut learning": when a predictor relies on shortcuts, i.e., spurious correlations between the inputs

This work was done when Fukun Yin was an intern at Tencent PCG.