Reproduce “Zero-Shot Temporal Resolution Domain Adaptation for Spiking Neural 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

Sanja Karilanova, Maxime Fabre, Emre Neftci, Ayça Özçelikkale

The authors adapt spiking-neural-network state-update parameters to a new time-bin width without target examples or changing synaptic weights. In simulated SHD 20-class spoken-digit classification, a non-recurrent adLIF model trained at bin 2 and tested at bin 1 reaches 89.5% accuracy with Integral adaptation, versus 53.0% with time-constant scaling. This 10-initialization result includes normalization adaptation and applies to this setting.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.