Differentiable Logic Gate Networks Run EEG Classification on Edge Devices Without Floating Point

arXiv·medium signal

Shyamal Y. Dharia, Stephen D. Smith and Camilo E. Valderrama (arXiv 2607.18149) replace conventional floating-point neural networks with differentiable logic gate networks for real-time EEG classification, targeting the arithmetic bottleneck on microcontroller-class hardware. Logic gate networks compile to boolean circuits, so inference is effectively free relative to matmul. This is a concrete data point for the broader thesis that edge inference will not be quantized transformers.