Image classification models using computer vision techniques have seen tremendous expansion in the past few decades. Many of these models, however, have been shown to produce biased results. For example, COMPAS, which predicts criminals likelihood of reoffending, has produced racially biased results. In our project, I will focus on the fairness aspect of biases and aim to develop a tool to improve fairness in the image classification domain. I hope to achieve this by using the Concept Bottleneck Model (CBM), under which each image first gets mapped into higher-level concepts (binary per concept) and then used as input for classification. With the recent development of tools like CLIP, the process of mapping images to concepts could be a more scalable and accurate process than before.

Knowledge with no application will disappear; work with no impact has limited meaning. I decided to participate in SuperUROP this year because I wanted to apply my skills and knowledge to produce tangible results. Moreover, I am extremely excited as my research will be at the intersection of technology and ethics. I hope to publish a paper by the end of the year and further the current understanding of this field by an inch and a half.