🚀 I'm incredibly excited to share that during the first year of my master's, I proposed an idea: using the concept of density estimation from generative models to assist distribution matching in imitation learning. This idea has led to two papers, each focusing on different aspects of imitation learning—offline and online.
The first paper, Diffusion Model-Augmented Behavioral Cloning (DBC), accepted at [ICML] Int'l Conference on Machine Learning 2024, enhances behavioral cloning by integrating diffusion models to model expert behaviors and guide policy learning. Our method demonstrates competitive performance in continuous control tasks like navigation, robot arm manipulation, and locomotion.
The second paper, Diffusion-Reward Adversarial Imitation Learning (DRAIL), accepted at NeurIPS 2024 🎉, integrates diffusion models into GAIL to produce more precise and smoother rewards for policy learning through a novel diffusion discriminative classifier.
📄 Paper: https://lnkd.in/gqPj-tb9
🌐 Project page: https://lnkd.in/g7jXEvB2
Stay tuned for the code!
I'm deeply grateful to my co-authors Shang-Fu Chen, Chun-Mao Lai, and Ming-Hao Hsu, and honored to collaborate with distinguished mentors like Prof. Ping-Chun Hsieh, Prof. Yu-Chiang Frank Wang, Dr. Min-Hung (Steve) Chen, and Prof. Shao-Hua Sun.
Now that I've graduated, I'm eager to dive into research on large language models (LLMs) or combining LLMs with reinforcement learning (RL).
#ICML2024 #NeurIPS2024 #ArtificialIntelligence #ReinforcementLearning #MachineLearning #LLM