Iteratively Refined Behavior Regularization for Offline Reinforcement Learning

Authors: Yi Ma, Jianye Hao, Xiaohan Hu, YAN ZHENG, Chenjun Xiao

NeurIPS 2024 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on the D4RL benchmark indicate that our method outperforms previous state-of-the-art baselines in most tasks, clearly demonstrate its superiority over behavior regularization.
Researcher Affiliation Collaboration Yi Ma1,2, Jianye Hao3,4 , Xiaohan Hu3, Yan Zheng3 Chenjun Xiao5 1School of Computer and Information Technology, Shanxi University, EMAIL 2Key Laboratory of Computational Intelligence and Chinese Information Processing of Ministry of Education 3College of Intelligence and Computing, Tianjin University EMAIL 4Noah s Ark Lab, Huawei 5The Chinese University of Hongkong, Shenzhen, EMAIL
Pseudocode Yes We give the pseudocode of both CPI and CPI-RE in Algorithm 1.
Open Source Code Yes Codes are provided in this link https://github.com/mamengyiyi/CPI.
Open Datasets Yes Experimental results on the D4RL benchmark indicate that our method outperforms previous state-of-the-art baselines in most tasks...
Dataset Splits No The paper states it uses D4RL benchmarks but does not explicitly provide details on training/validation/test dataset splits, such as percentages or specific sample counts for each split.
Hardware Specification Yes All experiments are run on a GeForce GTX 2080TI GPU.
Software Dependencies No The paper mentions using TD3+BC as a base for modifications and refers to a GitHub repository, but it does not specify explicit version numbers for programming languages (e.g., Python), deep learning frameworks (e.g., PyTorch), or other ancillary software libraries.
Experiment Setup Yes Table 3: CPI Hyperparameters and Table 4: Regularization parameter τ and weighting factor λ of CPI for all datasets detail the experimental setup, including specific hyperparameter values.