Waves Traveling Over A Map Of Visual Space Can Ignite Short-term Predictions Of Sensory Input

This article discusses a complex-valued neural network (cv-NN) that can process visual inputs ranging from simple stimuli to natural movies. This network is modified to include horizontal recurrent connections, which are thought to provide advantages over the standard FF architecture used in computer vision tasks. The article then explains how this understanding can be used to train recurrent complex-valued networks to process visual inputs and predict learned movies many frames into the future.