Reproduce “A Mixed-Precision CNN-LSTM Accelerator Featuring Long-Kernel Memory Reuse and Dynamic Quantization for Sleep Staging” 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
Xinyu Chen, Yaqian Xu, Yuhan Hou, Bowen Liu, Andrew G. Richardson, Xilin Liu
The authors present wearable EEG hardware for local sleep-stage inference. It reuses overlapping activations in long CNN kernels and combines static CNN quantization with dynamic recurrent-layer quantization and mixed precision. Across 31 folds, degradation from FP32 falls to 0.15% for five-class accuracy and 0.51% for N3 sensitivity; serpentine dataflow reduces long-kernel SRAM accesses 6.2×. This evaluates inference, not clinical benefit.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.