Conversion of Continuous-Valued Deep Networks to Efficient Event-Driven Networks for Image Classification

TLDR

Spiking neural networks can perform inference efficiently through sparse, event‑driven activation, and prior work has shown that simple CNNs can be converted to spiking form, though key operations such as max‑pooling, softmax, batch‑normalization, and Inception modules were missing; moreover, SNNs can trade classification error against the number of operations, unlike continuous networks that require a fixed operation count. This paper introduces spiking equivalents for these missing operations, enabling the conversion of almost any CNN architecture into a spiking network. The authors convert popular models—including VGG‑16 and Inception‑v3—into SNNs that achieve the best reported results on MNIST, CIFAR‑10, and ImageNet, and demonstrate that with only a few percentage points increase in error, the SNNs can cut operations by more than twice compared to the original CNNs.

Abstract

Spiking neural networks (SNNs) can potentially offer an efficient way of doing inference because the neurons in the networks are sparsely activated and computations are event-driven. Previous work showed that simple continuous-valued deep Convolutional Neural Networks (CNNs) can be converted into accurate spiking equivalents. These networks did not include certain common operations such as max-pooling, softmax, batch-normalization and Inception-modules. This paper presents spiking equivalents of these operations therefore allowing conversion of nearly arbitrary CNN architectures. We show conversion of popular CNN architectures, including VGG-16 and Inception-v3, into SNNs that produce the best results reported to date on MNIST, CIFAR-10 and the challenging ImageNet dataset. SNNs can trade off classification error rate against the number of available operations whereas deep continuous-valued neural networks require a fixed number of operations to achieve their classification error rate. From the examples of LeNet for MNIST and BinaryNet for CIFAR-10, we show that with an increase in error rate of a few percentage points, the SNNs can achieve more than 2x reductions in operations compared to the original CNNs. This highlights the potential of SNNs in particular when deployed on power-efficient neuromorphic spiking neuron chips, for use in embedded applications.