Hassain Alsayhah

Artificial Intelligence Specialization · Cohort 2023

FairGuard: Fair Image Classification Across Domains

Our proposed methodology enhances the robustness and fairness of our model by augmenting the training data. Specifically, we focus on the underrepresented class by introducing new multi-domain samples. This process serves a dual purpose: it not only minimizes the disparity in the sample distribution but also capitalizes on the benefits of multi-domain training. The end result is a noticeable improvement in the model's ability to accurately predict labels for the underrepresented group while simultaneously increasing its resilience against adversarial attacks. By bridging the gap between robustness and fairness, we have forged a path towards more equitable and reliable image classification.

See it in the program report (PDF)