Eva Yi Xie

Eva Yi  Xie
MIT EECS | Mason Undergraduate Research and Innovation Scholar
Advisor: Aleksander Madry
Department: Mathematics & Electrical Engineering and Computer Science
Years: 2022-2023
Research Project Title:

Text Conditional Image Generation for Identifying Subpopulation-Based Model Failure Modes

abstract:Previous studies on the critical role of image backgrounds in classification have largely been driven by a laborious counterfactual curation that does not provide quality data for arbitrary subpopulations. Thus, we utilize text-conditional image generation to emulate real-world distributions of images in a broad variety of contexts to efficiently provide high-quality data for investigation. Further, we evaluate to what extent text-conditional image generation can predict model performance in real-world subpopulations. We develop text conditional image generation into a powerful tool for identifying image classification model blind spots that originate from the model's dependency on the correlation between image backgrounds and labels in object recognition.
About:

SuperUROP provides me with an exciting way to conduct rigorous, meaningful research in Machine Learning. I am eager to develop modern machine learning toolkits to enable machine learning models to achieve high performance in a robust, responsible, and reliable way in the real world. I am eager to work in the Madry Lab around innovative thinkers.