Thanh Nguyen-Tang

Background

I am currently a postdoc at Johns Hopkins University (with Raman Arora). Prior to that, I did my PhD in Computer Science at The Applied AI Institute, Deakin University, Australia (Alfred Deakin Medal for Doctoral Theses). I did my M.Sc. in Computer Science at Ulsan National Institute of Science and Technology, South Korea. In my previous life, I studied Electronic and Communication Engineering (Talented Engineering Program) at Danang University of Science and Technology, Vietnam.

Research Interest

My research interest is the theory of learning and decision-making. I am currently working on: (offline) reinforcement learning, multi-agent learning, transfer learning, and large language models (LLMs). Some key questions that drive me currently are:

• How to learn optimally and efficiently from pre-collected experiences for decision making?

• How to learn and act optimally and efficiently in the presence of adaptive and/or strategic agents?

• How to optimally and efficiently transfer knowledge from one domain to another?

• What are LLMs such as transformer-based ones capable of computing, exactly?

• How to optimally and efficiently align LLMs with human values in any given domain?

Selected Publications by Topics

 — Make the world an \(\epsilon\)-better place

(see here for all publications or my Google Scholar)

Offline Reinforcement Learning with Large State Spaces

• On Sample-Efficient Offline Reinforcement Learning: Data Diversity, Posterior Sampling and Beyond. Thanh Nguyen-Tang, Raman Arora. Advances in Neural Information Processing Systems (NeurIPS), 2023.

• VIPeR: Provably Efficient Algorithm for Offline RL with Neural Function Approximation. Thanh Nguyen-Tang, Raman Arora. International Conference on Learning Representations (ICLR), 2023 (noble-top-25%- Spotlight) [ERRATUM].

• On Instance-Dependent Bounds for Offline Reinforcement Learning with Linear Function Approximation. Thanh Nguyen-Tang, Ming Yin, Sunil Gupta, Svetha Venkatesh, Raman Arora. AAAI Conference on Artificial Intelligence (AAAI), 2023.

• Offline Neural Contextual Bandits: Pessimism, Optimization and Generalization. Thanh Nguyen-Tang, Sunil Gupta, A.Tuan Nguyen, and Svetha Venkatesh. International Conference on Learning Representations (ICLR), 2022.

• On Sample Complexity of Offline Reinforcement Learning with Deep ReLU Networks in Besov Spaces. Thanh Nguyen-Tang, Sunil Gupta, Hung Tran-The, Svetha Venkatesh. Transactions on Machine Learning Research (TMLR), 2022.

Robust Reinforcement Learning

• Distributional Reinforcement Learning via Moment Matching. Thanh Nguyen-Tang, Sunil Gupta, Svetha Venkatesh. AAAI Conference on Artificial Intelligence (AAAI), 2021.

• Distributionally Robust Bayesian Quadrature Optimization. Thanh Nguyen-Tang, Sunil Gupta, Huong Ha, Santu Rana, Svetha Venkatesh. International Conference on Artificial Intelligence and Statistics (AISTATS), 2020.

Collaborative Multi-Agent Learning

• Optimistic Rates for Multi-Task Representation Learning. Austin Watkins, Enayat Ullah, Thanh Nguyen-Tang, Raman Arora. Advances in Neural Information Processing Systems (NeurIPS), 2023.

• Multi-Agent Learning with Heterogeneous Linear Contextual Bandits. Anh Do, Thanh Nguyen-Tang, Raman Arora. Advances in Neural Information Processing Systems (NeurIPS), 2023.

Selected Awards

• Alfred Deakin Medal for Doctoral Theses (for the most outstanding theses), 2022.

Teaching

• Co-instructor (with Raman Arora), Machine Learning: Advanced Topics: Foundations of Data-Driven Sequential Decision-Making Systems (CS 779), JHU, Spring 2024.

• Teaching RL Theory in our JHU ML reading group, Summer/Fall 2023. [notes]

• Guest lecturer (in bandits/reinforcement learning): Machine Learning (CS 475/675) Spring 2023, JHU. [notes]

• Teaching Assistant: Statistical Machine Learning, Fall 2017, UNIST; Engineering Programming I/II, Spring 2016, UNIST; Various advanced mathematics and engineering courses, 2012-2016, Vietnam.

* I participated in (and obtained a certificate of) Justice, Equity, Diversity, and Inclusion (JEDI) Training in the Classroom in March 2024 at JHU, as an effort to improve diversity in my future classes and research group.

Mentoring

• Ragja Palakkadavath (PhD student at Deakin University, out-of-distribution generalization)

• Thong Bach (independent researcher, self-supervised learning and domain adaptation)

• Anh Do (PhD student at JHU, bandit/reinforcement learning)

• Austin Watkins (PhD student at JHU, transfer learning and robustness)

• Cao Tinh (grad student at KU Leuven, learning theory in economic settings)

Others