Practical Boolean Backpropagation

TL;DR

This paper introduces a practical method for training neural networks entirely in Boolean algebra, bypassing traditional numeric computations, and demonstrates its feasibility through initial experiments.

Contribution

It presents the first practical approach for purely Boolean backpropagation in neural networks, focusing on a specific Boolean gate and operating directly in Boolean algebra.

Findings

Feasibility confirmed through initial experiments

Operates directly in Boolean algebra without numeric computations

Provides a hardware-efficient alternative to traditional neural network training

Abstract

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.