What are Neural Networks?

What are Neural Networks?

Neurons are traditionally linked with human brains. They are the basic working unit of the human brain. Neurons are specialized cells designed to transmit information to other cells and control our everyday functioning. Talking in short about how we process images, understands everything we see. We humans have a primary visual cortex, also known as V1, containing 140 million neurons, with 10 Billion connections between them. We have a series of Visual cortices V1, V2, V3, V4, V5 doing progressively more complex image processing. Well, all these seem too easy as we process everything subconsciously and we don’t even care. But things become very complex if we try to implement such a network. There are some genius Brains out there trying to imitate the network in Machines and study of this network is known as Artificial Neural Network.

How do Neural Networks work?

Scientists are using two basic Neuron to imitate the human Nervous system. They are Perceptron and Sigmoid Neuron. So, we will talk about these Neurons. Each neuron represents a grayscale value of the corresponding pixel ranging from 0 to 1 (0 – Black, 1- white). This number inside the neuron is called its “Activation”. Each Neuron gets lit up when its gray scale value is close to 1. The neurons combine together to form a layer. We have basic layers like input layer to feed our training samples, output layer to give us the output of the problem and few middle layers called Hidden layer to process the input. When we feed any training data into the input layer some processing is done by the neurons in the layer, the pattern is recorded and passed on to next layer. The next layer process the pattern and triggers neurons of the next layer. In this way Neurons triggering get us to the final layer of the Network and we get the output.

For eg. Consider a problem of recognizing hand-written digit. For this, we provide a handwritten digit image in the input layer lets take it to be 9. So the first layer recognizes 9’s pattern and lits up neurons in the second layers representing 9’s upper circle arcs and lower lines small parts. This second layer, in turn, lits up the neurons in the third layer representing 9’s upper circle and lower straight line. This two neuron will lit up neuron representing whole 9 in the output layer. This is how Neural network will recognize handwritten digit 9.

The basic terminology behind Neural networks is each neuron in the network takes several binary inputs and produces a single binary output. Each input has a weight associated with it. The neuron's output is 0 or 1 is determined by whether the weighted sum of inputs is less than some threshold value or not. The output is 1 if weighted sum is greater than the threshold value.

As an Analogy, suppose you want to go to college depends upon 3 input factors:-

1.If your bus arrives on time.

2. If your friend is coming.

3. If there is no rain.

Suppose the weight given to condition 1 is 3 and 2 is 4 and 3 is 6. That means your chances of going to college depends mostly on weather i.e if there are no rains. And you set the threshold value to 5. So, what the neurons do here is it checks for all the weighted sum which one is greater than your threshold. As here only condition 3 is greater than 5. Therefore network will output 1 or 0 depends on whether there will be rains or not. If there were 2 conditions whose weight were greater than the threshold then neuron would have considered both the inputs before outputting 0 or 1.

This was the basic idea behind Artificial Neural Network. I hope this helps. Finally, Thank you.