Reproduce “Let SSMs be ConvNets: State-space Modeling with Optimal Tensor Contractions” 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
Yan Ru Pei
Centaurus represents SSM blocks as tensor networks, enabling varied channel connectivity and reordered equivalent operations suited to tensor shapes. Optimized contractions reduce measured forward-plus-backward time on a single bottleneck block but do not establish whole-model training savings. On VoiceBank plus DEMAND denoising, the hybrid scores PESQ 3.12 versus 3.04 for the full variant; FRCRN scores 3.21, so the result is task- and comparison-specific.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.