Neuromorphic Computing Journey (Part 4)

Hello everyone to the 4th part of this series.

Its time to discuss little bit more abut Spiking Neural Network (SNN). SNN it third generation neural network that imitates the action of normal neurons in the brain. Here the standard non-linearity functions like Sigmoid, Relu functions were replaced by spiking functions. The first scientific model of SNN was proposed by Alan Hodgekin and Andrew Huxley in 1950s. This model discusses biological neurons’ action potentials initialization and propagation.

How does SNN work?

Well, the key difference between ANN and SNN is information propagation approach. Instead of working with continuously changing in time values used in ANNs, SNN operated with discrete events that occurs at certain point of time. SNN takes spike train as input and produces a spike train as the output.

At each timestamp, each neuron has some value that is analogous to the electrical potential of biological neurons. This value varies depending on the neuron model. If the value of neuron exceeds the threshold , the neuron send a single impulse (spike) to each downstream neuron connected to the initial one. After the spike the value of neuron drops instantly below the average. Here it experience the analogy of a biological neuron’s refractory period. With time the neuron’s value returns to its average.

Some Famous SNN Neuron Models

SNN has many different neuron models, but here we will discuss only the famous ones.

Leaky-Integrated-and-Fire (LIF) model

This is the most commonly used model because of its simplicity. In LIF, each cell has membrane potential (V), with capacitance © and a leaky channel which will allow the current to flow across the membrane with resistance (R). When the voltage exceeds the threshold value, the action potential (spike) is generated. After the spike, the voltage drops to the reset value.

Izhikevich neuron model

This model combines the biological plausibility of the Hodgkin-Huxley

model and computational efficiency of the integrate-and-fire neuron (LIF) model. This model reproduces the spiking and bursting behavior of known types of cortical neurons.

Some Advantages of SNNs

The major advantage of SNN is the temporal plasticity, reduced computational complexity, and ease of use in neural interface. In recent years, the popularity of SNNs and its models has increased and several algorithms have been developed for image classifications and object recognition using SNNs. The SNNs are suitable for a diverse range of applications related to computer vision and robotics such as classification, clustering, pattern recognition, etc.

We will discuss these advantages in detail in upcoming blogs.

Thanks for hanging in so far. I hope now you get the basic idea of what is Spiking Neural Network. In next blog I’ll discuss how we can do learning in SNNs.

Stay Tuned.