Enhancing Fairness in Deep Learning from Data and Model Perspectives

With the widespread use of deep learning models in practical applications, there are increasing concerns that these models could perpetuate or amplify societal biases and discrimination if not properly regulated. In this talk, I will discuss approaches to enhance the fairness of deep learning models from both data and model perspectives, highlighting their applications in computer vision and natural language processing tasks. Our framework is theoretically grounded, effective in balancing model performance and fairness, and computationally efficient.