Mirrored STDP Implements Autoencoder Learning in a Network of Spiking Neurons.

Kendra S Burbank,Lyle J Graham

doi:10.1371/journal.pcbi.1004566

Abstract

The autoencoder algorithm is a simple but powerful unsupervised method for training neural networks. Autoencoder networks can learn sparse distributed codes similar to those seen in cortical sensory areas such as visual area V1, but they can also be stacked to learn increasingly abstract representations. Several computational neuroscience models of sensory areas, including Olshausen & Field’s Sparse Coding algorithm, can be seen as autoencoder variants, and autoencoders have seen extensive use in the machine learning community. Despite their power and versatility, autoencoders have been difficult to implement in a biologically realistic fashion. The challenges include their need to calculate differences between two neuronal activities and their requirement for learning rules which lead to identical changes at feedforward and feedback connections. Here, we study a biologically realistic network of integrate-and-fire neurons with anatomical connectivity and synaptic plasticity that closely matches that observed in cortical sensory areas. Our choice of synaptic plasticity rules is inspired by recent experimental and theoretical results suggesting that learning at feedback connections may have a different form from learning at feedforward connections, and our results depend critically on this novel choice of plasticity rules. Specifically, we propose that plasticity rules at feedforward versus feedback connections are temporally opposed versions of spike-timing dependent plasticity (STDP), leading to a symmetric combined rule we call Mirrored STDP (mSTDP). We show that with mSTDP, our network follows a learning rule that approximately minimizes an autoencoder loss function. When trained with whitened natural image patches, the learned synaptic weights resemble the receptive fields seen in V1. Our results use realistic synaptic plasticity rules to show that the powerful autoencoder learning algorithm could be within the reach of real biological networks.

Highlights

Neurons in the brain’s sensory areas need to form useful internal representations of the external world
Autoencoder networks can successfully model neuronal responses in early sensory areas, and they are frequently used in machine learning for training deep neural networks
To perform the autoencoder algorithm, neurons must modify their incoming, feedforward synaptic connections as well as their outgoing, feedback synaptic connections—and the changes to both must depend on the errors the network makes when it tries to reconstruct its input

Summary

Introduction

Neurons in the brain’s sensory areas need to form useful internal representations of the external world. The preferred features are relatively simple for neurons in primary areas such as primary visual cortex (V1) and primary auditory cortex (A1), but increase in complexity, sparsity, abstractness, and size in higher brain areas. It is an intriguing possibility that the brain uses a similar mechanism to learn receptive fields in higher sensory areas as it does in the primary areas. If pairwise or higher order correlations in neuronal activity are present in one area, those correlations might be captured to form a more abstract representation in the area. We introduce a model for learning in a single area which we argue fulfills these requirements: it is biologically plausible while allowing varying levels of sparsity and producing representations that need not be uncorrelated

Methods

Results

Discussion

Conclusion