Baiting Zhu

10933 Rochester Ave, Los Angeles, CA 90024

I am an incoming MSCS student at Stanford University. I completed my undergraduate degree at UCLA majoring in Math of Computation (math & cs) with a minor in Statistics. My research interests are in the areas of Deep Reinforcement Learning and Generative Models. I am actively looking for potential research opportunities in Fall 2023!

Currently, I am a student researcher at Machine Intelligence (MINT) group where I am fortunate to be advised by Prof. Aditya Grover; and a research intern at Microsoft Research Asia (MSRA), ML/RL team where I am fortunate to be advised by Dr. Li Zhao and Pushi Zhang. Previously, I interned at Kuaishou Technology as an deep learning engineer where I built search models and managed data pipelines.

news

Mar 22, 2023	I will become a Stanfard MSCS student in Fall 2023.
Jan 21, 2023	Our work *Scaling Pareto-Efficient Decision Making via Offline Multi-Objective RL* is accepted to ICLR 2023!

selected publications

Scaling Pareto-Efficient Decision Making via Offline Multi-Objective RL

Baiting Zhu, Meihua Dang, and Aditya Grover

International Conference on Learning Representations (ICLR), 2023

Abs Bib PDF Code Poster Website

The goal of multi-objective reinforcement learning (MORL) is to learn policies that simultaneously optimize multiple competing objectives. In practice, an agent’s preferences over the objectives may not be known apriori, and hence, we require policies that can generalize to arbitrary preferences at test time. In this work, we propose a new data-driven setup for offline MORL, where we wish to learn a preference-agnostic policy agent using only a finite dataset of offline demonstrations of other agents and their preferences. The key contributions of this work are two-fold. First, we introduce D4MORL, (D)atasets for MORL that are specifically designed for offline settings. It contains 1.8 million annotated demonstrations obtained by rolling out reference policies that optimize for randomly sampled preferences on 6 MuJoCo environments with 2-3 objectives each. Second, we propose Pareto-Efficient Decision Agents (PEDA), a family of offline MORL algorithms that builds and extends Decision Transformers via a novel preference-and-return-conditioned policy. Empirically, we show that PEDA closely approximates the behavioral policy on the D4MORL benchmark and provides an excellent approximation of the Pareto-front with appropriate conditioning, as measured by the hypervolume and sparsity metrics.
@article{zhu2023paretoefficient, title = {Scaling Pareto-Efficient Decision Making via Offline Multi-Objective RL}, author = {Zhu, Baiting and Dang, Meihua and Grover, Aditya}, journal = {International Conference on Learning Representations (ICLR)}, year = {2023}, }