Phase-XOR Neural Networks: Finite-State, Reversible Perceptrons without Real-Valued Arithmetic
Published in Preprint on 2025
Recommended citation: Tsai, Leslie; Kumar, Jayanth; Jha, Prateek Chandra. "Phase-XOR Neural Networks: Finite-State, Reversible Perceptrons without Real-Valued Arithmetic." Preprint (2025). /files/papers/PXOR_Perceptron_jaykmr.pdf
This preprint presents Phase-XOR (PXOR) neural networks that operate on a four-valued phase logic alphabet {1, I, -1, -I} and replace real-valued weighted sums with cyclic phase composition. The work formalizes the PXOR perceptron, provides a deterministic learning rule with convergence guarantees, and demonstrates applications to parity learning, finite automata simulation, and phase-coded memory—highlighting reversible, hardware-efficient neural computation.