Exploiting similarity in data has been widely used in data-drive predictive analytics. In online recommendation systems like Netflix and Spotify that have both huge user bases and items to recommend, we can predict whether a user will like an item by looking at what similar users like. Despite such applications that use similarity-based nearest-neighbor approaches, there has been little theoretical development in understanding when and why these methods should work well. In this project, we aim to address these issues by building a common stack for large-scale nearest-neighbor inference with theoretical performance guarantees and that tackles a diverse range of applications such as time series classification, online collaborative filtering, and image segmentation

Since I started college I’ve been trying to combine my love for math and computer science and machine learning seems like a sweet spot for this. I took the undergraduate machine learning class the past semester and hope that this project together with future classes will help me advance in this field.