Mingxuan Li

@Mingxuan0422

PhD candidate @Columbia. Working on Causal Reinforcement Learning. Former RL researcher @BrownCSDept. @MATSprogram scholar. #CausalReinforcementLearning #CRL

mingxuan.me

Joined April 2019

572

Following

Followers

Replies Reposts Media

Pinned
Mingxuan Li
@Mingxuan0422
Sep 2
World model trained agents often fail in deployment. Identifying the real sim2real gap is hard. Our solution, Counterfactual Debugging, pinpoints the cause via causal attribution at 1M steps. 🧵(1/8)
1
Mingxuan Li
@Mingxuan0422
Jul 8
🎉 🎉 Excited to share our latest work on Confounding Robust Offline Reinforcement Learning (RL) presented at the International Conference on Machine Learning (ICML)! Our research highlights a critical issue: when an expert operates with rich state information that the learner
Mingxuan Li
@Mingxuan0422
Nov 28, 2025
look forward to seeing everyone in San Diego! Happy to chat if you are also interested in causal reinforcement learning!
Elias Bareinboim
@eliasbareinboim
Nov 28, 2025
Replying to @eliasbareinboim
4/8 “Confounding-Robust Deep Reinforcement Learning: A Causal Approach” (joint w/ @Mingxuan0422, @JunzheZhang12) Wed, 2 pm (#2511) Link: causalai.net/r132.pdf Off-policy RL can be fatally biased by unobserved confounders, which are pervasive in many real-world datasets. To
Mingxuan Li
@Mingxuan0422
Sep 19, 2025
Very excited to share that my recent work with @JunzheZhang12 @eliasbareinboim on confounding robust deep reinforcement learning has been accepted by NeurIPS 25! This is the first work that addresses unobserved confounding issue in offline reinforcement learning with a practical
Elias Bareinboim
@eliasbareinboim
Sep 19, 2025
Happy to share that it’s been an exciting season for our team with several NeurIPS papers accepted! These projects span causal representation learning, robust reinforcement learning, causal discovery, and interpretable modeling -- highlighting the development of causal
4
Mingxuan Li
@Mingxuan0422
Jul 12, 2025
Excited to share our new work "Automatic Reward Shaping from Confounded Offline Data" at ICML 25' with @JunzheZhang12, @eliasbareinboim! Reward shaping is a popular technique to tackle the sparse reward problem in RL by adding extra learning signals. But how could we design it in
5