binarized-neural-network-706aaeb6·1 events·first seen Aliases: Binarized Neural Network
Researchers evaluate Differentiable Logic Gate Networks (Diff-Logic) for real-time EEG classification on edge hardware, compiling models into pure Boolean circuits executable via bitwise CPU operations. Across four EEG datasets covering dementia detection and emotion recognition, Diff-Logic matches or outperforms MLP and Binarized Neural Network baselines while achieving up to 2.9× speedup and 14× smaller model size on a 7W Nvidia Jetson Orin Nano. Inference latency remains nearly constant as model scale increases 10×, making the approach attractive for resource-constrained brain-computer interfaces.