Zhang-Wei Hong
Research: My aim is to address practical search problems using tools for sequential decision-making such as reinforcement learning (RL) and the multi-arm bandit framework. I broadly interpret search challenges as discovery of nearly optimal solutions within an huge search space. Addressing these challenges is akin to strike a balance between exploration (deciding when to try the unknown) and exploitation (deciding when to leverage what's known). Thus, I believe that the sequential decision-making principles of balancing exploration and exploitation are crucial for devising effective solutions for search challenges. My research is focused on the following key areas:
- Sample Efficient RL: Given that resolving many search problems necessitates costly real-world interactions or the use of high-fidelity simulations, my early works are dedicated to creating RL algorithms that can learn effective policies more efficiently with minimal human intervention and less data. (References: NeurIPS'23, ICML'23, ICML'23, ICLR'23, ICLR'22, and ICLR'22)
- Domain-agnostic Exploration Strategies: At the core of solving search challenges is the capability to experiment with new solutions in an efficient manner, minimizing repetition and accurately estimating outcomes with fewer trials. Existing exploration methods often rely on estimating uncertainty through domain-specific features, necessitating significant human effort and trial and error. This limitation prevents us from applying single exploration strategy to many problems. My goal is to develop domain-agnostic exploration strategies. (References: NeurIPS'23)
- Applications: In addition to algorithms, I'm interested in addressing real-world problems by framing them as search challenges and applying my algorithms to these problems. My recent interests include applications in AI safety (e.g., red teaming), cybersecurity, software testing, and scientific endeavors. (References: ICLR'24)
Bio:
Zhang-Wei is a final year Ph.D. student in Electrical Engineering and Computer Science (EECS) at the Massachusetts Institute of Technology (MIT), advised by Prof. Pulkit Agrawal. In 2024, Zhang-Wei was awarded the Qualcomm Fellowship for North America. He completed both his B.S. and M.S. degrees at National Tsing Hua University, working closely with Prof. Chun-Yi Lee and Prof. Min Sun. Previously, Zhang-Wei had the opportunity to collaborate with Prof. Jan Peters at TU Darmstadt in Germany and worked at Preferred Networks.
|
|
|
Research intern
Feb. 2019 - Jun. 2019
Appier
Advisor: Prof. Min Sun
|
|
Red Teaming Language-Conditioned Robot Models via Vision Language Models
Sathwik Karnik*, Zhang-Wei Hong*, Nishant Abhangi*, Yen-Chen Lin, Tsun-Hsuan Wang, Pulkit Agrawal
Accepted as a workshop paper in NeurIPS Safe Generative AI Workshop 2024
Paper | Bibtex
|
|
Going Beyond Heuristics by Imposing Policy Improvement as a Constraint
Chi-Chang Lee, Zhang-Wei Hong*, Pulkit Agrawal
Accepted as a conference paper in NeurIPS 2024
Paper | Code | Bibtex
|
|
Random Latent Exploration for Deep Reinforcement Learning
Srinath Mahankali, Zhang-Wei Hong, Ayush Sekhari, Alexander Rakhlin, Pulkit Agrawal
Accepted as a conference paper in ICML 2024
Website |
Paper
|
|
Curiosity-driven Red-teaming for Large Language Models
Zhang-Wei Hong, Idan Shenfeld, Tsun-Hsuan Wang, Yung-Sung Chuang, Aldo Pareja, James R. Glass, Akash Srivastava, Pulkit Agrawal
Accepted as a conference paper in ICLR 2024
Paper |
Code |
News
Bibtex
|
|
Beyond Uniform Sampling: Offline Reinforcement Learning with Imbalanced Datasets
Zhang-Wei Hong, Aviral Kumar, Sathwik Karnik, Abhishek Bhandwaldar, Akash Srivastava, Joni Pajarinen, Romain Laroche, Abhishek Gupta, Pulkit Agrawal
Accepted as a conference paper in NeurIPS 2023
Paper |
Code |
|
|
Maximizing Velocity by Minimizing Energy
Srinath Mahankali*, Chi-Chang Lee*, Gabriel B. Margolis, Zhang-Wei Hong, Pulkit Agrawal
Accepted as a conference paper in ICRA 2024
Paper (coming soon)
|
|
TGRL: An Algorithm for Teacher Guided Reinforcement Learning
Idan Shenfeld, Zhang-Wei Hong, Aviv Tamar, Pulkit Agrawal
Accepted as a conference paper in ICML 2023
Paper |
Website |
Code |
|
|
Parallel Q-Learning: Scaling Off-policy Reinforcement Learning under Massively Parallel Simulation
Zechu Li*, Tao Chen*, Zhang-Wei Hong, Anurag Ajay, Pulkit Agrawal
Accepted as a conference paper in ICML 2023
Paper |
Code |
|
|
Harnessing Mixed Offline Reinforcement Learning Datasets via Trajectory Reweighting
Zhang-Wei Hong Pulkit Agrawal, Remi Tachet des Combes and Romain Laroche
Accepted as a conference paper in ICLR 2023
Paper |
Code |
|
|
Redeeming intrinsic rewards via constrainted optimization
Eric Chen*, Zhang-Wei Hong*, Joni Pajarinen and Pulkit Agrawal (* indiccates co-first author)
Accepted as a conference paper in NeurIPS 2022
Paper |
Website |
Code |
MIT News
|
|
Bilinear Value Networks for Multi-goal Reinforcement Learning
Zhang-Wei Hong*, Ge Yang*, and Pulkit Agrawal (* indiccates co-first author)
International Conference on Learning Representation (ICLR) 2022 - Conference paper
Paper | Code
|
|
Topological Experience Replay
Zhang-Wei Hong, Tao Chen, Yen-Chen Lin, Joni Pajarinen, and Pulkit Agrawal
International Conference on Learning Representation (ICLR) 2022 - Conference paper
Paper | Code
|
|
Stubborn: A Strong Baseline for Indoor Object Navigation
Haokuan Luo, Albert Yue, Zhang-Wei Hong , Pulkit Agrawal
IEEE/RSJ International Conference on Intelligent Robots and Systems 2022 - Conference paper
Paper |
Code
|
|
Reducing the Deployment-Time Inference Control Costs of Deep Reinforcement Learning Agents via an Asymmetric Architecture
Chin-Jui Chang, Yu-Wei Chu, Chao-Hsien Ting, Hao-Kang Liu, Zhang-Wei Hong, and Chun-Yi Lee
International Conference on Robotics and Automation (ICRA) 2021 - Conference paper
Paper
|
|
Periodic Intra-Ensemble Knowledge Distillation for Reinforcement Learning
Zhang-Wei Hong, Prabhat Nagarajan, and Guilherme Maeda
European Conference on Machine Learning (ECML) 2021 - Conference paper
Paper
|
|
Adversarial Active Exploration for Inverse Dynamics Model Learning
Zhang-Wei Hong, Tsu-Jui Fu, Tzu-Yun Shann, Yi-Hsiang Chang, and Chun-Yi Lee
Conference on Robot Learning (CoRL) 2019 - Oral
Paper | Project
|
|
Diversity-Driven Exploration Strategy for Deep Reinforcement Learning
Zhang-Wei Hong, Tzu-Yun Shann, Shih-Yang Su, Yi-Hsiang Chang, Tsu-Jui Fu, and Chun-Yi Lee
Neural Information Processing Systems (NeurIPS) 2018 - Poster
International Conference on Representation Learning (ICLR) Workshop 2018
Paper | Project
|
|
Virtual-to-Real: Learning to Control in Visual Semantic Segmentation
Zhang-Wei Hong, Chen Yu-Ming, Shih-Yang Su, Tzu-Yun Shann, Yi-Hsiang Chang, Hsuan-Kung Yang, Brian Hsi-Lin Ho, Chih-Chieh Tu, Yueh-Chuan Chang, Tsu-Ching Hsiao, Hsin-Wei Hsiao, Sih-Pin Lai, and Chun-Yi Lee
International Joint Conference on Artificial Intelligence (IJCAI) 2018 - Oral
Paper | Project
|
|
Deep Policy Inference Q-Network for Multi-Agent Systems
Zhang-Wei Hong, Shih-Yang Su, Tzu-Yun Shann, Yi-Hsiang Chang, and Chun-Yi Lee
International Conference On Autonomous Agents and Multi-Agent Systems (AAMAS) 2018 - Oral
Paper
|
|
Tactics of adversarial attack on deep reinforcement learning agents
Yen-Chen Lin, Zhang-Wei Hong, Yuan-Hong Liao, Meng-Li Shih, Ming-Yu Liu, and Min Sun
International Joint Conference on Artificial Intelligence (IJCAI) 2018 - Poster
Paper | Project
|
|