Cortical code and connectivity
作者单位：Tsinghua University
会议名称：《中国神经科学学会第十二届全国学术会议》
会议日期：2017年
学科分类：0710[理学-生物学] 07[理学] 071006[理学-神经生物学]
摘 要：Recurrent connections were discovered more than a decade ago in primary visual cortex. But its functional role still remains debated. In this paper, we trained a convolutional neural network with recurrent connections on natural images. We show that the results connectivity and coding pattern resembles those discovered in biological experiments. Furthermore, we reveal the surprising fact that recurrent connections in this model increases robustness rather than performance.