|
Hi, I am a 4th year PhD student at Machine Learning Department,
Carnegie Mellon University. I am advise by Aarti Singh and
Drew Bagnell.
I am also closely working with Wen Sun. I am partially supported by
Two Sigma PhD fellowship.
My email is yudas at andrew dot cmu dot edu. Google Scholar / CV / Github / Twitter |
|
|
I am interested in the theory, science and application of interactive decision-making. My current study focuses on when and how we can achieve efficient and robust learning, from thinking about the three foundations of interactive decision-making: data, environment and interaction protocol. I am also interested in the application of principled decision-making algorithms in large-scale real-world applications, such as generative models and robotics. |
|
|
|
To Distill or Decide? Understanding the Algorithmic Trade-off in Partially Observable Reinforcement Learning
Yuda Song, Dhruv Rohatgi, Aarti Singh, J Andrew Bagnell Neurips, 2025 |
|
Outcome-based Exploration for LLM Reasoning
Yuda Song, Julia Kempe, Remi Munos Neurips ARLET workshop, 2025 |
|
Accelerating Unbiased LLM Evaluation via Synthetic Feedback
Zhaoyi Zhou, Yuda Song, Andrea Zanette ICML, 2025 |
|
Mind the Gap: Examining the Self-Improvement Capabilities of Large Language Models
Yuda Song, Hanlin Zhang, Udaya Ghai, Carson Eisenach, Sham M. Kakade, Dean Foster ICLR, 2025 |
|
The Importance of Online Data: Understanding Preference Fine-Tuning Through the Lens of Coverage
Yuda Song, Gokul Swamy, Aarti Singh, J. Andrew Bagnell, Wen Sun NeurIPS, 2024. We prove that offline contrastive-based method (e.g., DPO) requires a stronger coverage property than online RL-based method (e.g., RLHF). We propose Hybrid Preference Optimization to combine the benefits of both offline and online methods. |
|
Hybrid Reinforcement Learning from Offline Observation Alone
Yuda Song, J. Andrew Bagnell, Aarti Singh ICML, 2024. |
|
Rich-Observation Reinforcement Learning with Continuous Latent Dynamics
Yuda Song, Lili Wu, Dylan J. Foster, Akshay Krishnamurthy ICML, 2024. We introduce a new theoretical framework, RichCLD (Rich-Observation RL with Continuous Latent Dynamics), in which the agent performs control based on high-dimensional observations, but the environment is governed by low-dimensional latent states and Lipschitz continuous dynamics. |
|
Provable Benefits of Representational Transfer in Reinforcement Learning
(alphabetical order) Alekh Agarwal, Yuda Song, Wen Sun, Kaiwen Wang, Mengdi Wang, Xuezhou Zhang COLT, 2023. [code] We prove the benefit of representation learning on diverse source environments which allows efficient learning on the source environment with the learned representation under the low-rank MDPs setting. |
|
The Virtues of Laziness in Model-based RL: A Unified Objective and Algorithms
Anirudh Vemula, Yuda Song, Aarti Singh, J. Andrew Bagnell, Sanjiban Choudhury ICML, 2023. [code] |
|
Hybrid RL: Using Both Offline and Online Data Can Make RL Efficient
Yuda Song*, Yifei Zhou*, Ayush Sekhari, J. Andrew Bagnell, Akshay Krishnamurthy, Wen Sun ICLR, 2023. [code] [Talk at RL theory seminars] Combining online data and offline data can solve RL with both statistical and computation efficiency. Experiments on Montezuma's Revenge reveals that hybrid RL works much better than pure online RL and pure offline RL. |
|
Representation Learning for General-sum Low-rank Markov Games
Chengzhuo Ni, Yuda Song, Xuezhou Zhang, Chi Jin, Mengdi Wang ICLR, 2023. |
|
Efficient Reinforcement Learning in Block MDPs: A Model-free Representation Learning Approach
Xuezhou Zhang, Yuda Song, Masatoshi Uehara, Mengdi Wang, Alekh Agarwal, Wen Sun ICML, 2022. [code] [Talk at RL theory seminars] An efficient rich-observation RL algorithm that learns to decode from rich observations to latent states (via adversarial training), while balancing exploration and exploitation. |
|
No-regret Model-Based Meta RL for Personalized Navigation
Yuda Song, Ye Yuan, Wen Sun, Kris Kitani L4DC, 2022. |
|
Transform2Act: Learning a Transform-and-Control Policy for Efficient Agent Design
Ye Yuan, Yuda Song, Zhengyi Luo, Wen Sun, Kris Kitani ICLR, 2022. [Project Page] |
|
PC-MLP: Model-based Reinforcement Learning with Policy Cover Guided Exploration
Yuda Song, Wen Sun ICML, 2021 [code] A simple provably efficient model-based algorithm that achieves competitive performance in both dense reward continuous control tasks and sparse reward control tasks that require efficient exploration. |
|
Provably Efficient Model-based Policy Adaptation
Yuda Song, Aditi Mavalankar, Wen Sun, Sicun Gao ICML, 2020 [Project Page] [code] We study Sim-to-Real/policy transfer/policy adaptation under a model-based framework resulting an algorithm that enjoyes strong theoretical guarantees and excellent empirical performance. |
|
|
|
|
|
Source code from here |
