Reproduce “Threshold Neuron: A Brain-inspired Artificial Neuron for Efficient On-device Inference” 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
Zihao Zheng, Yuanchun Li, Jiayu Chen, Peng Zhou, Xiang Chen, Yunxin Liu
Threshold-Net replaces conventional weighted-sum neurons with units that compare inputs against learned thresholds and subtract the threshold from passing inputs before aggregation. On CIFAR-10, Threshold-ResNet18-FW256 reaches 88.53% accuracy, versus 85.64% for DenseShift. Synthesized kernel circuits show 3.89×–4.33× lower area and 7.51×–8.19× lower circuit energy; a four-layer FPGA prototype reports 2.52× power savings and 1.75× speedup, not GPU inference speedups.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.