Stochastic Resonance in Ternary Neural Computation: Brownian Noise as a Stabilising Substrate for Diffusion-Based Intelligence

Noise is conventionally treated as an obstacle to robust computation. We demonstrate the inverse: that Brownian motion injected at the input layer of a ternary neural network functions as a stabilising substrate, analogous to the forward diffusion process in generative models. We derive a closed-form expression for the optimal noise amplitude as a function of system dimensionality and ternary encoding depth. Empirically, we validate this framework on bioelectric time-series data collected from 500 bovine subjects over a 90-day pilot, demonstrating that Brownian noise injection improves prediction horizon stability by 340% over deterministic baselines.