Reproduce “Common-Mode Errors Limit Low-Timestep Deep Spiking Q-Networks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zijie Xu, Bingrui Guo, Yiding Sun, Yiting Dong, Zhile Yang, Zhaofei Yu
CMC-DSQN addresses low-timestep spiking Q-network learning by correcting shared action-value error during training with an auxiliary artificial neural network (ANN). The networks are jointly trained, but the ANN is removed at inference, leaving greedy action selection to the spiking network. On MiniAtar at two timesteps, it achieves 90.30% average performance ratio versus 72.46% for CaRe-BN, averaged across five games rather than guaranteed game by game.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.