What is a Neural Network?
Introduction to Neural Network:
Neural networks were first proposed in 1944 by Warren McCullough and Walter Pitts, two University of Chicago researchers who moved to MIT in 1952 as founding members. Neural Networks is the heart of Deep Learning and a subset of Machine Learning which has contributed a lot in the fields of Artificial Intelligence. The aim of introducing neural networks was to create a replica of network in the human brain and make the Artificial Neural Network function the same way as human brain network works like making decisions, etc
A neural Network consist of millions of nodes called as neurons which are densely interconnected layer by layer. A neuron in one layer is connected to every other neuron in the next layer. Each neuron accepts the input from the previous layer and gives the output to the next layer. Each neuron in the network is associated with weights which are either randomly initialized or initialized by using different weight initialization techniques.
How the Neural Network works?
Firstly the inputs or the training data is passed to the input layer. Then these inputs gets multiplied with the weights of each of the neuron and gets added with the bias. Here inputs can be text, numbers, images ,speech ,etc.
The formula of how the inputs and weights are multiplied and added with the bias is given below,
∑wixi + bias = w1x1 + w2x2 + w3x3 +…….wnxn+ bias
Where,
w is the weight associated with each neuron,
x is the input passed to the neuron.
After calculation it gives a value to the node which is compared with the threshold value. If this value is above the threshold value then the neuron(node) is activated otherwise not. The activation of the neuron depends upon the activation function. We will be learning activation function in the upcoming blog. The values of the activated neurons are passed to the hidden layer and again the same processing is done in the hidden layer and further the output of the hidden layer is passed to the output layer in the form of probabilities. Considering two neurons as per the above diagram Fig 1,the output layer has two neurons so here the neuron which gets the value with highest probability is considered to be the output. This is how the neural network model is trained and the process of calculating the output on the basis of probabilities is called the testing process. If the predicted output is matching with the actual output then this means that our model is trained properly. If the predicted output is not matching with the actual output then again the weights in the neural networks are adjusted till the model is trained and gives the correct predicted output by using backpropagation algorithm. We will know about the backpropagation in detail in the next blogs.
Further the prediction is done on a new set of data and similarly the output is calculated on an totally unknown but similar data. By unknown and similar it means that ,if you are predicting between a cat and a dog then the images used for the training of the model are different from those used for prediction but the image data will be of the same category that is cat or a dog.
If you are getting confused between training ,testing and predicting then lets clear it out with a small example. Consider a scenario of a JEE aspirant in which Student goes to the coaching and the teachers in coaching class trains the student in solving different types of questions. A test is been conducted by the coaching class in which the questions are quite similar or exact to the questions on which the student was trained by the teacher or the questions on which the student has done his/her self study. So this justifies the testing part of the student (model).
Now the training and testing of the data is done. Now remains the prediction capacity of the model.
Here the prediction of the model means ,the students final JEE exam in which the type of questions is bit uncertain or unknown. And the student (model) is been evaluated on the basis of final JEE exam.
Hope you got an basic idea of Neural Network.
Thank You!
