Laixi Shi

Postdoctoral Fellow, Caltech CMS

Contact: laixis@caltech.edu

I am currently a postdoctoral fellow in the Dept. of Computing + Mathematical Sciences (CMS) at California Institute of Technology, hosted by Prof. Adam Wierman and Prof. Eric Mazumdar. My research interests focus on exploring new frontiers of data-driven solutions in the real-world, establishing theoretical foundations and developing faithful practical solutions.

Previously, I obtained my Ph.D in Electrical and Computer Engineering at Carnegie Mellon University in 2023, supervised by Prof. Yuejie Chi. I received my bachelor's degree from Electronic Engineering, Tsinghua University in 2018. I interned at Google Research (Brain Team) in Paris and Mountain View, as well as at Mitsubishi Electric Research Laboratories (MERL).

I received 5 Rising Star awards in electrical engineering and computer science, machine learning, signal processing, and (computational) data science, and 2 Ph.D. Presidential Fellowships. My Ph.D. thesis won the CMU ECE A.G. Milnes Award (2024).

Decision-Making (RL) Theory: Robustness, Data Efficiency, and Scalability: Thesis as a summary; online RL, offline RL(I and II), robust single-agent RL(I and II), and robust multi-agent RL (I and II)
Decision-Making (RL) Practice: Benchmark, Control, Robotics, Resource Allocation: Benchmark --- Robust Gymnasium, control and robotics (I and II), autonomous driving, and curriculum RL.
Data-Driven Optimization for Inverse Problems: Sensing and Processing: Multi-channel blind deconvolution, image stitching, indoor localization, strain measurement, hand gesture recognition.

[2024/9/25] Three papers got accepted by NeurIPS 2024. Thanks to my wonderful collaborators.

[2024/7/12] One paper got accepted by JMLR.

[2024/5/10] My thesis won the CMU ECE A.G. Milnes Award, which is awarded to a graduating ECE PhD student for the PhD thesis work judged to be of the highest quality and which has had or is likely to have significant impact in his or her field. Credits to Prof. Yuejie Chi and all collaborators!

[2024/5/14] Two papers got accepted by RLC 2024. Thanks to my wonderful collaborators.

[2024/5/1] Two papers got accepted by ICML 2024. Thanks to my wonderful collaborators.

[2024/11/01] Speaker at the 44th Southern California Control Workshop at USC.

[2024/10/27-30] Invited Speaker at the Asilomar Conference on Signals, Systems, and Computers.

[2024/10/24-25] EECS Rising Stars Workshop at MIT.

[2024/10/20-23] Session Chair at 2024 INFORMS Annual Meeting.

[2024/10/9-11] Invited to The 2024 Young Researchers Workshop at Cornell.

[2024/06/10-14] Instructor in Reinforcement Learning Bootcamp at Caltech!.

[2024/04/19] 2024 ISyE Junior Researcher Workshop in Georgia Tech.

Breaking the Curse of Multiagency in Robust Multi-Agent Reinforcement Learning

Laixi Shi*, Jingchu Gai*, Eric Mazumdar, Yuejie Chi, Adam Wierman.
In submission

[Arxiv]

Robust Gymnasium: A Unified Modular Benchmark for Robust Reinforcement Learning

Shangding Gu*, Laixi Shi*, Muning Wen, Ming Jin, Eric Mazumdar, Yuejie Chi, Adam Wierman, Costas Spanos
In submission

[Github]

Sample-Efficient Robust Multi-Agent Reinforcement Learning in the Face of Environmental Uncertainty

Laixi Shi, Eric Mazumdar, Yuejie Chi, Adam Wierman.
International Conference on Machine Learning (ICML), 2024

[Arxiv]

The Curious Price of Distributional Robustness in Reinforcement Learning with a Generative Model

Laixi Shi, Gen Li, Yuting Wei, Yuxin Chen, Matthieu Geist, Yuejie Chi.
Under submission to Operations Research
Conference on Neural Information Processing Systems (NeurIPS), 2023

[Paper] [Slides]

Settling the Sample Complexity of Model-Based Offline Reinforcement Learning

Gen Li, Laixi Shi, Yuxin Chen, Yuejie Chi, Yuting Wei.
The Annals of Statistics, 2024.

[Paper]

Distributionally Robust Model-Based Offline Reinforcement Learning with Near-Optimal Sample Complexity

Laixi Shi and Yuejie Chi.
Journal of Machine Learning Research (JMLR), 2024

[Paper] [Slides]

Pessimistic Q-Learning for Offline Reinforcement Learning: Towards Optimal Sample Complexity

Laixi Shi, Gen Li, Yuting Wei, Yuxin Chen, Yuejie Chi.
International Conference on Machine Learning (ICML), 2022.

[Paper]

Breaking the Sample Complexity Barrier to Regret-Optimal Model-Free Reinforcement Learning

Gen Li, Laixi Shi, Yuxin Chen, Yuejie Chi
Information and Inference: A Journal of the IMA, 2023.
NeurIPS Spotlight, 2021

[Paper] [Arxiv] [Talk] [Slides] [Poster]

Manifold Gradient Descent Solves Multi-channel Sparse Blind Deconvolution Provably and Efficiently

Laixi Shi, Yuejie Chi
IEEE Transactions on Information Theory, 2021.
International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2020.

[Paper] [Slides] [Poster]

Micro Hand Gesture Recognition System Using Ultrasonic Active Sensing

Yu Sang, Laixi Shi, Yimin Liu
IEEE Access, 2018.

[Paper] [Video]

Non-Wearable Three-Dimensional Display and Interaction Cube Based on Multi-Screen

Tsinghua University, Beijing, China.

We designed a real-time virtual reality cubic interaction metaphor, arranged 5 LED panels into a cube to create a 3D object illusion, utilizing binocular vision to track user’s eyes and rendering the perspective of each screen to fit the user’s eyes positions.

Collaborative Distributed Formation System based on Cars/Unmanned Aerial Vehicle

Tsinghua University, Beijing, China

Collaborated with Han Zhang, Wenhao Ding, and Tuopu Wen. We developed a multiple robot collaborative localization and navigation system using a self-made mobile robot platform. The robots in our system are equipped with a wireless distributed localization system based on UWB(Ultra Wide Band), which enables them automatically to form a pre-defined formation with mutual localization information. (Image from my collaborator Wenhao Ding)

Air Hockey Robot

Tsinghua University, Beijing, China

We built a robot to play air-hockey game with human, which is a popular table game. This robot consists of a bird-view camera (capturing the images of moving ball) and an executive arm (a striking device with two degrees of freedom). (Image from my collaborator Wenhao Ding)