Research Assistant, Rutgers University

I worked with Rutgers professor Sudarsun Kannan on the project titled "An Examination on the Efficacy of Systems-Level Techniques for Optimizing Recommendation Models." As a case study, we optimized Facebook's Deep Learning Recommendation Model (DLRM), focusing on the data preprocessing step.

Highlights

• Cut data-preprocessing runtime by more than 80% using memory-mapped I/O, madvise access hints, and multiprocessing across cores.
• Won a university CS research award for the results and the presentation.