§ Research Intern

I have been fortunate enough to play with some interesting and challenging optimization tasks for training neural recommendation models at an industrial scale. These recommenders, along with their embedding tables, easily exceed trillion bytes in size. Training at this scale creates many interesting stochastic and distributed optimization problems, and calls for usecase-dependent method development.

• Research Intern, Facebook (Meta), Ads Ranking
  • Date: May 2021 - Nov 2021.
  • Mentor: Dr. Xiaohan Wei, Dhruv Choudhary.
  • Developed a method that is computation and memory efficient for embedding learning. Work summarized in this paper (ICLR 2022).

• Research Intern, Bytedance, AML
  • Date: May 2020 - Aug 2020.
  • Mentor: Dr. Chong Wang.
  • Developed training techniques that prevent degenerate learning in the deep retrieval model.

§ Industry Collaboration

I have also explored the topic of multi-agent reinforcement learning in a year-long collaboration with Ford.

• Ford Motor Company
  • Date: Sep 2020 - Dec 2021.
  • See here for a milestone presentation of the project.
  • A method that promotes the efficiency & robustness of deep multi-agent RL algorithms (NeurIPS 2023).