What is Neural Network and How They Work?

How Neural Network is connected with the working of the Human Brain?

Anyone that got curious about Machine Learning(Artificial Intelligence) and was fascinated about teaching a machine to recognize things and patterns must have tried coding the AI and faced a concept called Neural Network. If you are that person who is exploring this concept then this article will help you to get a grasp of that.

A neuron is a biological term that is used to explain a nerve cell that helps in conducting signals to the brain. We can say that they are the basic unit of the brain. When we are learning new things the connections between the neurons change so that we could perform the things that we have learned faster and more efficiently. Similarly, an artificial neuron is a function that takes one or more than one input and returns one output. The output of the function depends upon the inputs called weight. We use the weights and put them against a condition and the result decides the output for example if the sum of the weight is greater than 6 then the output will be 6 else 5. It can be seen as a function that decides by weighing the evidence.

How Neural Network Functions?

So how does a simple function help us in teaching a machine? The answer is that a single function may not seem like a big deal but when we make a network with layers of these neuron functions it kind of simulate our mind. The layers of neurons that do not directly take the initial input or give final output are called hidden layers because the user does not interact with them. To make it more human-like we introduced bias to these neurons which are nothing but a default value in the function that is seen as a weight so even without any input weight a neuron function contains some weight. Just like the human brain for eg. Let’s say that you have to decide whether to go to a party where your friend will come. So when you are making a decision of whether or not to go. you will be weighing your options like there are your friends but the weather is not good and so on but there is always some bias like in this case you could have a bias against or towards going to the party that will affect your decision of going. Just like that, we give a bias to the functions.

These neuron functions are called perceptrons and they work pretty good but to make a system learn we would have to change those biases a little by little to change the incorrect result but in the network made with these types of neurons. A little change in on input of the function can drastically affect the result of the network therefore we use a new type of neurons called sigmoid neurons where a slight change in the network will change the output only a little bit so the learning is better with these neuron functions. Sigmoid neurons use sigmoid functions which use exponentials, but we also use different types of values and functions to change them out with respect to the input like root mean square and many others.

These type of networks in which the input from one layer of the neurons are going to the next layer and so on till the output neurons is called a feedforward networks. The networks which use the output from the neurons of the next layer as their input (feedback results) are called recurrent neural networks. These are normally used for specific complex problems.

Readout our more similar blogs at i4consulting.