Complexity and diversity in sparse code priors improve receptive field characterization of Macaque V1 neurons.

Ziniu Wu,Yimeng Zhang,Harold Rockwell,Shiming Tang,Tai Sing Lee

doi:10.1371/journal.pcbi.1009528

Abstract

System identification techniques—projection pursuit regression models (PPRs) and convolutional neural networks (CNNs)—provide state-of-the-art performance in predicting visual cortical neurons’ responses to arbitrary input stimuli. However, the constituent kernels recovered by these methods are often noisy and lack coherent structure, making it difficult to understand the underlying component features of a neuron’s receptive field. In this paper, we show that using a dictionary of diverse kernels with complex shapes learned from natural scenes based on efficient coding theory, as the front-end for PPRs and CNNs can improve their performance in neuronal response prediction as well as algorithmic data efficiency and convergence speed. Extensive experimental results also indicate that these sparse-code kernels provide important information on the component features of a neuron’s receptive field. In addition, we find that models with the complex-shaped sparse code front-end are significantly better than models with a standard orientation-selective Gabor filter front-end for modeling V1 neurons that have been found to exhibit complex pattern selectivity. We show that the relative performance difference due to these two front-ends can be used to produce a sensitive metric for detecting complex selectivity in V1 neurons.

Highlights

The neural code of neurons in the primary visual cortex has been investigated for decades [1,2,3]
We propose an improvement of these standard methods by using a set of neural codes learned from natural scene images based on the convolutional sparse coding theory as priors or the front-end for these methods. We found that this approach improves the model performance in predicting neural responses with less data and with faster convergence for fitting, and allows a possible interpretation of the constituents of the receptive fields in terms of the dictionary learned from natural scenes
We investigate the idea that a set of basis filters learned from natural scene images based on convolutional sparse coding could be used to improve Projection pursuit regression (PPR) and convolutional neural networks (CNNs) in neuronal receptive fields (RFs) recovery tasks

Summary

Introduction

The neural code of neurons in the primary visual cortex has been investigated for decades [1,2,3]. Our experiments show that CNNs with a diverse complex-shape code front-end achieve the state-ofthe-art performance in neuronal response prediction, but are more data-efficient to train and faster in convergence. This result suggests that this more diverse set of basis functions learned from natural scenes might provide a better approximation of the underlying components of the neurons’ RFs than the noisy filters, characterized by white-noise like power spectrum and lacking in coherent structures, often learned by CNN trained from scratch

Methods

Results

Discussion

Conclusion