Deep Learning Predicts Correlation between a Functional Signature of Higher Visual Areas and Sparse Firing of Neurons.

Chengxu Zhuang,Yulong Wang,Daniel Yamins,Xiaolin Hu

doi:10.3389/fncom.2017.00100

Chengxu Zhuang, Yulong Wang + Show 2 more

Open Access

https://doi.org/10.3389/fncom.2017.00100

Copy DOI

Abstract

Visual information in the visual cortex is processed in a hierarchical manner. Recent studies show that higher visual areas, such as V2, V3, and V4, respond more vigorously to images with naturalistic higher-order statistics than to images lacking them. This property is a functional signature of higher areas, as it is much weaker or even absent in the primary visual cortex (V1). However, the mechanism underlying this signature remains elusive. We studied this problem using computational models. In several typical hierarchical visual models including the AlexNet, VggNet, and SHMAX, this signature was found to be prominent in higher layers but much weaker in lower layers. By changing both the model structure and experimental settings, we found that the signature strongly correlated with sparse firing of units in higher layers but not with any other factors, including model structure, training algorithm (supervised or unsupervised), receptive field size, and property of training stimuli. The results suggest an important role of sparse neuronal activity underlying this special feature of higher visual areas.

Highlights

After a complex visual pattern enters the visual system of mammals, the pattern undergoes different processing stages
Many neurons in the primary visual cortex (V1) are sensitive to edges (Hubel and Wiesel, 1962, 1968), some neurons in the visual area V2 are sensitive to line conjunctions or corners (Hegde and Van Essen, 2000; Ito and Komatsu, 2004), and some neurons in the inferior temporal cortex are sensitive to the whole pattern, such as faces or cars (Kanwisher et al, 1997; Epstein and Kanwisher, 1998; Gauthier et al, 2000)
Along the ventral visual pathway, higher areas, including areas V2 and V4, play more important roles than V1 for the perception of natural texture images (Freeman et al, 2013; Okazawa et al, 2015), but the mechanism underpinning this functional signature of the higher areas is unclear

Summary

Introduction

After a complex visual pattern enters the visual system of mammals, the pattern undergoes different processing stages. Many neurons in the primary visual cortex (V1) are sensitive to edges (Hubel and Wiesel, 1962, 1968), some neurons in the visual area V2 are sensitive to line conjunctions or corners (Hegde and Van Essen, 2000; Ito and Komatsu, 2004), and some neurons in the inferior temporal cortex are sensitive to the whole pattern, such as faces or cars (Kanwisher et al, 1997; Epstein and Kanwisher, 1998; Gauthier et al, 2000). The differences among the simple response properties of neurons in various areas are not always prominent and robust. The neural responses to many artificial stimuli in V2 are largely similar to those in V1 (Peterhans and Vonderheydt, 1989; Hegde and Van Essen, 2000; Lee and Nguyen, 2001).

Objectives

Methods

Results

Conclusion