UL Neural Net

Cross-Domain Spectral Neural Network

UL Neural Net is a 72,264-parameter pure Python neural network built from tensors up — no PyTorch, no TensorFlow. It uses spectral regularization derived from Monster group constants (via the Unity Lang project) and trains across 18 scientific domains including physics, quantum mechanics, chemistry, biology, fluid dynamics, and topology. Features include 3-phase curriculum learning, distributed P2P training, BSV-anchored state checkpoints, and a unified training pipeline that normalizes raw scientific data into a standard 32-feature format. The neural network serves as the empirical testbed for the observations described in the "Unconventional Quantum Paradigms in Computation" paper series. Published January 2026 by Bryan W. Daugherty, Gregory Ward, and Shawn Ryan.

Training Domains