Stochastic Computing for Edge Inference

***What if every multiplication in a vision transformer was a single logic gate?***

We treat **stochastic computing (SC)** as a *dense adaptive quantizer*: every multiply collapses to one AND/XNOR gate, and precision is set by bit-stream length — any positive integer, on the same hardware, at runtime. Validated end-to-end across **classification, detection, segmentation, image generation, and visual world-model planning.**