CFlowNets: Continuous Control with Generative Flow Networks

Authors: Yinchuan Li, Shuang Luo, Haozhi Wang, Jianye HAO

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, experimental results on continuous control tasks demonstrate the performance advantages of CFlow Nets compared to many reinforcement learning methods, especially regarding exploration ability. (...) To demonstrate the effectiveness of the proposed CFlow Nets, we conduct experiments on several continuous control tasks with sparse rewards, including Point-Robot-Sparse, Reacher-Goal-Sparse, and Swimmer-Sparse.
Researcher Affiliation Collaboration Yinchuan Li1, Shuang Luo2 , Haozhi Wang3, Jianye Hao1,3 1Huawei Noah s Ark Lab, Beijing, China 2Zhejiang University, Huangzhou, China 3Tianjin University, Tianjin, China
Pseudocode Yes The pseudocode is provided in Appendix C. (Algorithm 1 Generative Continuous Flow Networks (CFlow Nets) Algorithm is presented in Appendix C.)
Open Source Code Yes The codes are available at http://gitee.com/mindspore/models/tree/master/research/gflownets/cflownets
Open Datasets Yes To demonstrate the effectiveness of the proposed CFlow Nets, we conduct experiments on several continuous control tasks with sparse rewards, including Point-Robot-Sparse, Reacher-Goal-Sparse, and Swimmer-Sparse. (...) Both Reacher-Goal-Sparse and Swimmer-Sparse are adapted from Open AI Gym s Mu Jo Co environment.
Dataset Splits No The paper does not provide specific percentages or sample counts for training, validation, and test splits. It mentions collecting '10000 trajectories' but not how they are partitioned for different phases.
Hardware Specification No The paper does not explicitly describe the hardware used to run its experiments, such as specific GPU or CPU models.
Software Dependencies No The paper mentions software components like 'Adam' optimizer but does not provide specific version numbers for any key software libraries, frameworks, or environments used (e.g., PyTorch version, Gym version).
Experiment Setup Yes We provide the hyper-parameters of all compared methods under different environments in Table 1, Table 2, Table 3, Table 4, and Table 5. (...) As for Total Timesteps , Start Traning Timestep , Max Episode Length , Actor Network Hidden Layers , Critic Network Hidden Layers , Optimizer , Learning Rate , and Discount Factor , we set them the same for all algorithms for a fair comparison.