Winfree Oscillatory Neural Network Achieves Competitive Performance on Vision and Reasoning Benchmarks

A new neural architecture called the Winfree Oscillatory Neural Network (WONN) has been proposed, drawing on synchronization dynamics from neuroscience. Unlike prior machine learning approaches that limited oscillatory models to object discovery, WONN generalizes Winfree dynamics to operate on a torus (S^1)^d, enabling structured oscillatory interactions. The architecture combines phase-based inductive biases with flexible hierarchical mechanisms, which can be either fixed trigonometric mappings or learnable neural networks. Evaluated on CIFAR, ImageNet, Maze-hard, and Sudoku, WONN achieved competitive or superior performance on both image recognition and complex reasoning tasks. The work appears on arXiv with identifier 2605.20922.