On Emergences of Non-Classical Statistical Characteristics in Classical Neural Networks

Researchers introduce Non-Classical Network (NCnet), a classical neural architecture that exhibits quantum-like statistical behaviors through gradient competitions between neurons. The study reveals that multi-task neural networks can develop non-local correlations without explicit communication, providing new insights into deep learning training dynamics.

• →NCnet architecture demonstrates non-classical statistical behaviors similar to quantum mechanics in classical neural networks.
• →Non-classicality arises from gradient competitions of hidden-layer neurons shared across multiple tasks.
• →Task heads can implicitly sense other tasks' training through local loss oscillations, creating non-local correlations.
• →The S statistic serves as a potential indicator of model generalization performance and training dynamics.
• →Non-classical statistics offer a novel framework for understanding internal interactions in deep networks.