"Achieving Fairness and Better Trade-off in Dermatological Disease Diagnosis with Enhanced Balanced Incremental Distillation Netw"

"As deep learning technology is deeply empowering the practice of dermatological disease diagnosis, its diagnostic accuracy has steadily improved. Simultaneously, ensuring fairness in decision-making has emerged as a critical issue that requires urgent attention. However, existing research indicates improving fairness often comes at the cost of reduced diagnostic performance. To mitigate unfair discrimination against under-represented demographic groups while achieving a better trade-off between diagnostic accuracy and fairness, we propose an enhanced balanced incremental distillation network (EBID-Net). Specifically, aided by balanced memory, representative demographic groups are designed to assist underrepresented groups in learning knowledge, which is incrementally trained by integrating the distributions of different groups.  Additionally, we incorporate global information into the contextual attention mechanism to capture correlated interactions between features across different spatial locations, so as to obtain more robust feature representations during the incremental learning. Experimental results show that our network outperforms other methods in terms of fairness criteria and the trade-off between fairness and diagnostic accuracy."