Evolving Connectivity for Recurrent Spiking Neural Networks

Authors: Guan Wang, Yuhao Sun, Sijie Cheng, Sen Song

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate EC on a series of standard robotic locomotion tasks, where it achieves comparable performance with deep neural networks and outperforms gradient-trained RSNNs, even solving the complex 17-Do F humanoid task.
Researcher Affiliation Academia Guan Wang1, 2 , Yuhao Sun2, 3 , Sijie Cheng1, 4, Sen Song2, 3 1Deptartment of Computer Science and Technology, Tsinghua University 2Laboratory of Brain and Intelligence, Tsinghua University 3Department of Biomedical Engineering, Tsinghua University 4Institute for AI Industry Research (AIR), Tsinghua University
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes Our code is publicly available at https://github.com/imoneoi/Evolving Connectivity .
Open Datasets Yes Tasks. We focus on three robotic locomotion tasks in our experiments, Humanoid, Walker2d, and Hopper, as they are commonly used for sequential decision-making problems in the reinforcement learning domain [Brockman et al., 2016, Freeman et al., 2021].
Dataset Splits No The paper focuses on evaluating performance on reinforcement learning tasks rather than using explicit train/validation/test dataset splits. There are no mentions of specific validation sets or how data was partitioned for validation.
Hardware Specification Yes As a result, the training process achieves over 180, 000 frames per second on a single NVIDIA TITAN RTX GPU. [...] 8x NVIDIA Titan RTX GPU (24GB VRAM) 2x Intel(R) Xeon(R) Silver 4110 CPU 252GB Memory
Software Dependencies No Our EC framework and all baselines are implemented using the JAX library [Bradbury et al., 2018] and just-in-time compiled with the Brax physics simulator [Freeman et al., 2021] for efficient GPU execution. The paper mentions JAX and Brax but does not provide specific version numbers for these libraries.
Experiment Setup Yes Each experiment s result is averaged over 3 independent seeds, with the standard deviation displayed as a shaded area. For detailed information on hyperparameters and hardware specifications, please refer to Appendix G. [...] Appendix G.2 Hyperparameters lists specific values in Table 3, 4, 5, and 6.