Unraveling neural coding of dynamic natural visual scenes via convolutional recurrent neural networks

Yajing Zheng, Shanshan Jia, Zhaofei Yu, Jian K. Liu, Tiejun Huang

Abstract: Visual neuroscience is an immensely popular topic in AI, such that numerous methodologies developed in visual computing have broad applications and provide inspiration for other domains. The retina is one of the best-understood examples in neuroscience for visual computing. Here, we use retinal data to demonstrate how to use deep-learning models to encode dynamic visual scenes. The proposed models demonstrate that recurrence plays a critical role in encoding complex natural scenes and learning the biological computational underpinning of the neural circuits.