An artificial neural network is composed of a large number of Highly interconnected processing units those processing units are called as neurons those neurons are connected so that we can solve a particular problem like a human being for example the human being learn a particular concept or anything by experience for example initially with some examples he may learn or he may get some knowledge but as and when he has been shown with more number of examples the human being will learn or get more knowledge similar to that one as and when you go on giving more and more number of example to artificial neural network it will go on getting more and more knowledge in this case or it will learn more and more an artificial neural network is configured for a specific application or to solve a particular problem that is to solve a Spam classification phase recognition pattern recognition or similar to such kind of any application that is you cannot solve all the problems using one artificial neural network a particular artificial neural network can solve one problem at a time that is the one more very important part we need to remember but brain is brain works in such a way that it will try to solve or it will solve every problem over here each neuron is connected with other neuron with a connection Link in artificial neural network As I said earlier, each computation unit is called as a neuron and the network contains multiple numbers of neurons. They are connected with one another, and that connection is done with a link. Each and every link is associated with a particular weight. Whenever we give a signal to that particular neuron, a neuron that input signal and the weight will be transferred to the next neuron with the help of this particular link. In this case, this information, that is the weight, whatever we have assigned to that particular link or whatever it is learned by that particular artificial neural network, is used to solve a particular problem. In this case, the artificial neural network is characterized by their ability to learn, recall, and generalize the training patterns or the data. Similar to the human brain, the way the human brain works, with the same note, the artificial neural network works like they will learn and then they recall and generalize the training pattern. Over here, they have the capability to model the networks of original neurons as found in the brain. That's the reason these artificial neural networks are called as the processing units of artificial neural networks are called as the neurons that are also called as the artificial neurons. In this case, each neuron has an internal state of its own. As I said earlier, there will be multiple numbers of neurons in a network, and each of them has their own internal state. That internal state is called as the activation level. That activation level is transmitted to other neurons with the help of other links. In this case, we will take a very simple example to understand the simple artificial neural network. In this artificial neural network, we have two layers: the first one is called the input layer, and there is another layer called the output layer. These are the minimum two layers every artificial neural network has. In the input layer, we have two neurons that are denoted by X1 and X2, and Y is the output neuron. In this case, the output layer has one neuron, and the input layer has two neurons. So, in total, we have three neurons in this particular network. This X1 and X2 are the inputs to this particular neuron, and Y is the output of this particular output neuron. As I said earlier, each of these neurons is connected with a link. In this case, X1 and Y are connected with this link, X2 and Y are connected with this link, and each and every link is associated with some weight. Initially, we have assigned some random weight to each and every link, and we perform some computation here. Finally, we will get this particular output. Once you get this particular output, or once the termination condition is satisfied, whatever the weight we have assigned to these two weights (W1 and W2), that is the final one. Otherwise, if the termination condition is not satisfied, we have to go back and modify these particular weights. To modify these particular weights, we have different algorithms like gradient descent, backpropagation, and so on. We have to apply and update these particular weights in this case. Now, what kind of computation takes place in each and every neuron?For example, at this particular neuron, why the two kind of things take place or every neuron does these two things? That is, it performs something called a summation, and then it performs something called, it will apply something called an activation function over here on the top of that particular summation. That is, summation is nothing but the sum of the product of input and weight. Here, that is nothing but you can say that Y_in is equivalent to X1 * W1 + X2 * W2. In this case, that is, X1 is input, W1 is the weight, X2 is the input, and W2 is the weight. In this case, now this is the first thing performed at this particular output layer, and then once you get this particular Y_in, we need to perform something called, or we need to apply something called an activation function. So, once you apply this particular activation function on the top of Y, we will get the output here. As I said earlier, if you reach this particular termination condition, these are the final weights. Otherwise, we have to go back and modify this particular weight unless and until we reach the termination condition here. And one more very important thing is, this activation function is not a fixed one. Based on our requirement, we can apply some different activation functions like step function, sine function, sigmoid, ReLU, and so on. There are a lot of activation functions out there. Based on our requirement, we need to apply this particular sigma, this activation function, on the top of this particular summation term so that we can get the output here. Once you get the output, you can use this particular thing for making the decision, like whether we have to terminate this particular thing or we need to go and update these particular weights here.