This project aims to optimize handling sparse data in large-scale applications, specifically in recommendation engines and topic modeling machine learning. We aim to optimize Sparse Matrix operations like sparse matrix vector multiplication (SpVM) or sparse matrix dense matrix multiplication (SpMM) using cache optimizations, data prefetching, and parallelization. First, C++ implementations of these solutions will be hand optimized. These implementations will then be integrated into larger scale applications that involve large data sets. Contributing to the sparse data problem would motivate further applications such as faster and less memory-heavy graph algorithms, which has implications in ML algorithms (graph neural networks, convolutional neural networks), modeling, and graph analytics.

The UROP has been an integral part of my MIT experience, and I’m excited about SuperUROP to capstone that part of my undergraduate research career. I also look forward to working with my group this year and applying the skills that I have accumulated at MIT onto a project with real world implications. SuperUROP will also develop my communication skills so I can better present my ideas in my future career.