Abstract:Artificial neural networks (ANNs) have led to landmark changes in many fields, but they still differ significantly from the mechanisms of real biological neural networks and face problems such as high computing costs, excessive computing power, and so on. Spiking neural networks (SNNs) provide a new approach combined with brain-like science to improve the computational energy efficiency, computational architecture, and biological credibility of current deep learning applications. In the early stage of development, its poor performance hindered the application of SNNs in real-world scenarios. In recent years, SNNs have made great progress in computational performance and practicability compared with the earlier research results, and are continuously producing significant results. Although there are already many pieces of literature on SNNs, there is still a lack of comprehensive review on SNNs from the perspective of improving performance and practicality as well as incorporating the latest research results. Starting from this issue, this paper elaborates on SNNs along the complete usage process of SNNs including network construction, data processing, model training, development, and deployment, aiming to provide more comprehensive and practical guidance to promote the development of SNNs. Therefore, the connotation and development status of SNN computing is reviewed systematically and comprehensively from four aspects: composition structure, data set, learning algorithm, software/hardware development platform. Then the development characteristics of SNNs in intelligent computing are summarized, the current challenges of SNNs are discussed and the future development directions are also prospected. Our research shows that in the fields of machine learning and intelligent computing, SNNs have comparable network scale and performance to ANNs and the ability to challenge large datasets and a variety of tasks. The advantages of SNNs over ANNs in terms of energy efficiency and spatial-temporal data processing have been more fully exploited. And the development of programming and deployment tools has lowered the threshold for the use of SNNs. SNNs show a broad development prospect for brain-like computing.

A Survey of Stochastic Computing Neural Networks for Machine Learning Applications

A Survey of Neuromorphic Computing Based on Spiking Neural Networks

A Survey of Intelligent Chip Design Research Based on Spiking Neural Networks

Stochastic Computing Hardware Design and Optimization for Convolutional Neutral Networks

SC-DCNN: Highly-Scalable Deep Convolutional Neural Network using Stochastic Computing

Hardware-aware Neural Architecture Search for Stochastic Computing-Based Neural Networks on Tiny Devices

An Energy-Efficient and Noise-Tolerant Recurrent Neural Network Using Stochastic Computing

BitSNNs: Revisiting Energy-efficient Spiking Neural Networks

Survey on Energy-Efficient Deep Neural Networks for Computer Vision

An Energy-Efficient Bayesian Neural Network Implementation Using Stochastic Computing Method

Stochastic Computation Based Hybrid Artificial Neural Networks and Spike Neural Network

From Multipliers to Integrators: A Survey of Stochastic Computing Primitives

A Survey of Stochastic Computing in Energy-Efficient DNNs On-Edge

Not Your Father’s Stochastic Computing (SC)! Efficient Yet Accurate End-to-End SC Accelerator Design

Stochastic Computing for Neuromorphic Applications

The Promise and Challenge of Stochastic Computing

Towards Budget-Driven Hardware Optimization for Deep Convolutional Neural Networks Using Stochastic Computing

Stochastic Computing Convolution Neural Network Architecture Reinvented For Highly Efficient Artificial Intelligence Workload on Field Programmable Gate Array

Reconfigurable Neural Synaptic Plasticity-Based Stochastic Deep Neural Network Computing

Parallel Convolutional Neural Network (CNN) Accelerators Based on Stochastic Computing

A Review of Computing with Spiking Neural Networks