Practical Boolean Backpropagation
arXiv:2505.03791v1
Originality Incremental advance
AI Analysis
This addresses the need for hardware-efficient neural networks, but it appears incremental as it builds on existing Boolean network research.
The paper tackles the problem of training purely Boolean neural networks for hardware efficiency by presenting a Boolean backpropagation method based on a specific gate, with initial experiments confirming feasibility.
Boolean neural networks offer hardware-efficient alternatives to real-valued models. While quantization is common, purely Boolean training remains underexplored. We present a practical method for purely Boolean backpropagation for networks based on a single specific gate we chose, operating directly in Boolean algebra involving no numerics. Initial experiments confirm its feasibility.