BwNN (Black and White Neural Network) is a concept for training neural networks without floating‑point arithmetic. It represents numerical magnitude as structured binary dimensions rather than scalar weights, encoding knowledge as comparisons and compositions to enable logic‑native learning while still capturing “shades of grey.” This note establishes academic priority for the concept and outlines the high‑level hypotheses and validation plan, with implementation details to follow.
Zhi An (Wed,) studied this question.