Introduction to Artificial Neural Networks (ANN) and Spiking Neural Networks (SNN)

In this chapter, we introduce the reader to the basics of machine learning (ML)/neural networks (NN). Since neuromorphic computing is about implementing this kind of algorithms (and its variations) while consuming low energy, it’s important to know these ML/NN basics. Based on whether information is encoded in NN as floating point numbers or spikes, NNs can be divided into two types: artificial neural networks (ANN) and spiking neural networks (SNN), with ANN comprising majority of the neural networks used in the current day for ML. In this chapter, we discuss architecture, forward computation, and training of various kinds of ANNs in detail. We also discuss how ANNs can be implemented through crossbar arrays of non-volatile memory (NVM) synaptic devices (spintronic device being one of them). Then, we delve into SNNs and discuss similar aspects of SNN, while also throwing light on the biological inspiration of SNNs.