Jesse Zhang

CS PhD Student

University of Southern California

Hi! I am a third-year PhD student at USC, advised by Professors Jesse Thomason and Erdem Biyik at USC and Joseph J. Lim at KAIST on deep reinforcement learning for robotics. I completed my undergrad at UC Berkeley, where I worked with Professors Sergey Levine and Dinesh Jayaraman. I have interned at NAVER AI Labs and Horizon Robotics.

I am currently interning this summer at AWS doing research in Rasool Fakoor’s team.

I am interested in improving generalization and sample-efficiency of reinforcement learning algorithms by injecting inductive biases (e.g., programs), incorporating large offline datasets (e.g., through offline RL), and guiding agents with external knowledge (e.g., large language models).

Interests

Deep RL
Offline Skill Learning
Program Synthesis
Robotics

Education

CS PhD, 2020 - ?

USC
BA in Computer Science, 2016 - 2020

UC Berkeley

Recent News

[5/31/23] Started an internship at AWS research this summer!

[11/22/22] SPRINT is accepted as a spotlight talk at CoRL 2022 LangRob Workshop!

[4/22/22] I have received the ICLR 2022 highlighted reviewer award (top 8%)!

[12/15/21] Gave a talk at the AIPlans Workshop at NeurIPS 2021!

Featured Publications

Jesse Zhang, Karl Pertsch, Jiahui Zhang, Joseph J. Lim

June, 2023 ArXiv Preprint

SPRINT: Scalable Policy Pre-Training via Language Instruction Relabeling

Pre-training robot policies with a rich set of skills can substantially accelerate the learning of downstream tasks. Prior works have defined pre-training tasks via natural language instructions, but doing so requires tedious human annotation of hundreds of thousands of instructions. Thus, we propose SPRINT, a scalable offline policy pre-training approach which substantially reduces the human effort needed for pre-training a diverse set of skills. Our method uses two core ideas to automatically expand a base set of pre-training tasks: instruction relabeling via large language models and cross-trajectory skill chaining through offline reinforcement learning. As a result, SPRINT pre-training equips robots with a much richer repertoire of skills. Experimental results in a household simulator and on a real robot kitchen manipulation task show that SPRINT leads to substantially faster learning of new long-horizon tasks than previous pre-training approaches.

Dweep Trivedi, Jesse Zhang, Shao-Hua Sun, Joseph J. Lim

August, 2021

Learning to Synthesize Programs as Interpretable and Generalizable Policies

NeurIPS 2021

Jesse Zhang, Haonan Yu, Wei Xu

January, 2021 ICLR 2021

Hierarchical Reinforcement Learning by Discovering Intrinsic Options

ICLR 2021

Avi Singh, Albert Yu, Jonathan Yang, Jesse Zhang, Aviral Kumar, Sergey Levine

November, 2020 CoRL 2020

COG: Connecting New Skills to Past Experience with Offline Reinforcement Learning

CoRL 2020

Recent Publications

Quickly discover relevant content by filtering publications.

Machine Learning on Visibility Graph Features Discriminates the Cognitive Event-Related Potentials of Patients with Early Alzheimer's Disease from Healthy Aging

We present a framework for electroencephalography (EEG)-based classification between patients with Alzheimer’s Disease (AD) and …

Jesse Zhang, Jiangyi Xia, Xin Liu, John Olichney

Hierarchical Neural Program Synthesis

Preprint

Linghan Zhong, Ryan Lindeborg, Jesse Zhang, Joseph J. Lim, Shao-Hua Sun

Minimum Description Length Skills for Accelerated Reinforcement Learning

ICLR 2021 Self-Supervised Learning for RL Workshop

Jesse Zhang, Karl Pertsch, Jiefan Yang, Joseph J Lim

Minimum Description Length Skills for Accelerated Reinforcement Learning

Cautious Adaptation For Reinforcement Learning in Safety-Critical Settings

ICML 2020

Jesse Zhang, Brian Cheung, Chelsea Finn, Sergey Levine, Dinesh Jayaraman

Cautious Adaptation For Reinforcement Learning in Safety-Critical Settings

MobilityNet: Towards A Public Dataset For Multi-Modal Mobility Research

ICLR 2020 Climate Change AI Workshop

Kalyanaraman Shankari, Jonathan Fuerst, Mauricio Fadel Argerich, Eleftherios Avramidis, Jesse Zhang

MobilityNet: Towards A Public Dataset For Multi-Modal Mobility Research

See all publications

Teaching

CSCI 566: Deep learning (GSI @ USC, Spring 2023):
· Held office hours, advised 4 deep learning project teams, configured course to use Gradescope

CSCI 360: Intro to AI (GSI @ USC, Spring 2022):
· Held in-class discussion sections, office hours, created new written homework assignments, wrote test questions, and reconfigured course grading structure to use Gradescope

CSCI 566: Deep Learning (GSI @ USC, Fall 2020):
· Gave 2 lectures (RL for Robotics), advised 6 deep learning project teams, and held office hours

CS 188: Intro to AI (undergraduate GSI @ UC Berkeley, Fall 2019):
· Lead a discussion section and held office hours
· Received a teaching rating of 4.75/5, 0.42 above the department average

CS 170: Intro to CS Theory (course reader @ UC Berkeley, Spring 2019):
· Held office hours and volunteered to write problems for and help run extra sections on difficult material**

Service

Reviewing:
· Serving/Served as a reviewer for ICML 2023, ICLR 2023, CoRL 2022, ICML 2022, ICLR 2022 (highlighted reviewer award, top 8%), NeurIPS 2021 (outstanding reviewer award, top 8%), CoRL 2021, ICLR 2021, ICLR SSL-RL Workshop 2021, IEEE ITSC 2019

Irvine ML Organization:
· Serving as a principal advisor for Irvine ML Org, which teaches high schoolers about ML in Irvine, CA