Achieve fairness without demographics for dermatological disease diagnosis

Ching-Hao Chiu, Yu-Jen Chen, Yawen Wu, Yiyu Shi, Tsung-Yi Ho

Abstract: Highlights•Existing debias method involve using demographic data during training process.•Privacy issues may restrict the use of demographic data.•Feature entanglement regularization achieves fairness without demographic data.•Accurate masking disease features can improve both fairness and accuracy.