What is Artificial Neural Network and How it mimics the Human Brain?
This is the article for the newcomers in the field of Machine Learning. In this article I will try to give you a brief intuition about Artificial Neural Networks.
Artificial Network Network is one of the Machine Learning algorithms that mimics the human brain.It is built on the same concept as how human brain does work.At the end of this article we will also learn about Perceptron.
Human brain, the command center for the human nervous system is one of the most mysterious things of the world.The history of study of human brain is thousands of years old.The core component of human nervous system is the neuron.Around 100 billion neurons create a neuron’s network by connecting with each other.Information are transmitted from one neuron to another by electrical impulses and chemical signals.
Neurons are excitable cell.All neurons contain dendrites,a cell body(contains nucleus) and an axon.The dendrites look like the branches of a tree which collect information. The axon looks like a long tail which transmits information or messages between different cells. Soma(cell body)process the input signal to generate electric pulse and chemical signals.Two different neurons communicate with each other by sending neurotransmitter(a chemical) across the synapses between axons and dendrites of adjacent neurons when the neuron meets its threshold level.
An Artificial Neural Network in machine learning is an information processing technique that uses the same concept of biological neural networks but not identical to it.In the 1950's, the first step towards the simulate of hypothetical neural network was led by Nathanial Rochester from the IBM research laboratories.But unfortunately, his first attempt to do so was failed. In 1958 a psychologist Frank Rosenblatt invented the first Artificial Neural Network called Perceptron.The pattern-matching and learning capabilities of ANNs (Artificial Neural Networks) helped to solve many difficult problems.By the late 1980s, many real world institutes started to use ANNs for various purposes. Nowadays Artificial Neural Networks are used in language processing,machine translation,language generation, Stock market prediction , cancer cell analysis, Chatbots, aircraft fault detection and many more.In brief,Artificial Neural Networks bring a great revolution in the field of Artificial Intelligence.
ANN consists of three basic components -
1.Input Layer : contains the inputs.
2.Hidden Layers : contains one or multiple layers of neurons.
3.Output Layer : here the outputs are generating.
ANNs complete its learning procedure by following two strategies :
1.Forward propagation
2.Backward propagation
In forward propagation, different computational methods are performed to generate output.Backward propagation are used to minimize the error between the actual output and the generated output by the ANN model by updating the weights.We will learn about the weights and biological representation of an ANN in below.
Now let’s look at the figure 3. It is a schematic of an ANN inspired by biological neural networks where x1,x2,x3,….,xn represents the inputs and y1,y2,y3,..,ym represents the outputs. Inputs may be any integer or float or doubles values. Inputs are taken by dendrites in ANN and these inputs are passed to the neurons. Here neurons acts like a Soma(cell body) that process the inputs.
In ANN, we can add weight values to every dendrite. It is the most important factor in ANN. It is similar to slope in linear regression. Each neurons sum up the multiplied values of inputs with weights. This multiplication of inputs with the corresponding dendrite weight values converts the inputs to an impactful output.Then different activation function is used to produce the output. It is like the threshold voltage in neuron that helps the neuron to generate a meaningful and useful output signal. Outputs are transmitted across the axons.
Let’s elaborate this concept with a perception(a simple Artificial Neural Network with one neuron).
Let the input values are x1=1,x2=0,x3=1. The weights are w1=0.2,w2=0.4,w3=0.5. Let the bias value b=0 in this case.Bias values are used to adjust the weighted sum of the input values with the output values. Here in figure 4,the weighted sum of the input values is,
x1*w1 +x2*w2 +x3*w3 = 1*0.2 + 0*0.4 +1*0.5 =0.7
Now,this weighted sum is passed through a activation function. Lets say the activation function is sigmoid. So,the result after passing through the activation function is,
y = f(0.7) = 0.6682
So, the output of the figure 4 is 0.6682.
Figure 4 shows a single neuron. Many neurons like this by connecting with each other creates a complex Artificial Neural Network.
