Neural Architecture Inspired

We introduce Ring Convolution Networks (RCN), a novel neural network architecture where each weight exists in a quantum-inspired superposition of states. Unlike traditional neural networks with fixed weights, RCN implements a "ring structure" where each weight has mirror states that create an adaptive receptive field. Our experiments demonstrate that RCN achieves 19.8% improvement over standard random projections on MNIST, while showing no improvement on random or smooth synthetic data. This selective advantage suggests RCN acts as an implicit feature detector particularly suited for structured data like images. We provide theoretical analysis linking our architecture to quantum computing principles and demonstrate how classical systems can emulate quantum superposition effects. Code available at: https://github.com/Akbar1992A/ring-convolution-networks