Adaptive linear neuron, also known as Adeline Network, in artificial neural networks, the computational unit with the linear activation function are called as the linear units. A network with a single linear unit is called as adaptive linear neuron, also known as Adeline Network. In this case, because we use a linear unit in Adeline, the relationship between input and output is linear. In this case, Adeline uses bipolar activation function for its input signal and for its output signals over here. This is how the architecture of Adeline looks like. If you want to understand this particular architecture, I will take three examples. Let us say that 1, 2, and 3 are the three examples, and X1, X2, and Xn are the features, and then I will consider Y_n as the calculated output over here. Now, if you look at this particular architecture, X0 is the bias here that is always equivalent to one, so that's the reason I have not written it. X1, X2, Xn are the features with respect to each and every example over here, and then we need to calculate Y_n. To calculate this particular Y_n, first, we will assign some random weights over here. The random bias and the weights are W1, W2, Wn in this case. These particular inputs are passed to this particular summation function or the activation function using these particular weights, and the Y_n, that is the value of this particular neuron, is calculated using this particular equation: B + summation of X_i * W_i, where B is the bias, X_i is the input, and W_i is the weight over here. Now, if you consider this particular first example, these are the values that will be given as an input, and then we will calculate this particular value over here, that is, Y_n. It will be calculated. Now, once you calculate this particular Y_n, what we do is we will go back and then update this particular weights here as well as bias in this case. Now, how to update these particular weights and biases? We use this particular equation: W_i_new = W_i_old + Alpha * (T - Y_n) * X_i. Here, this is the old weight, whatever we have before we do this particular updation. Alpha is the learning rate, T is the target for that particular example, Y_n will be the calculated value over here, and X_i is the input. So, with the help of this particular equation, we will update W1, W2, till WN over here. The bias is updated with the help of this equation: Bias_new = Bias_old + Alpha * (T - Y_n), where T is the target, and Y_n is the calculated value over here. Now, once you're done with this particular first example, we will give the second example as an input over here. This one will be given as an input, and then we will again calculate this particular Y_n. Now, once you calculate this particular Y_n, again, we will go back and update these particular weights. The same thing will be done for the third input also. Now, once you complete this particular third input, the meaning of this one is we have completed one epoch here. Now, before we go to this particular second epoch (that is, epoch number two, and so on), what we do is we will calculate the total error of this particular epoch here. After giving the first input or the first example as an input, we do one thing: we will calculate the error with respect to it. That is, E1 is equivalent to the summation of (T - Y_n)². So, whatever the Y we got in the first iteration will be considered to calculate this particular E1. When you give the second input, we will calculate E2 here, and when you give the third input, we will calculate E3 here. The sum of all these values, what do you think, is nothing but the total error of that particular epoch. Once you calculate the total error of that particular epoch, we will decide whether to go with the second epoch or we should not go here. For example, if the acceptable error is E, and the calculated error of this particular epoch is E_t (that is the total error), and the calculated error is less than this particular expected error, we can stop over here. But if the calculated error is greater than this particular expected error, we need to go to the second epoch and then we need to do this particular computation again over there. So that, we will try to reduce this particular total error here. Once the total error becomes less than the expected error, then we can stop this particular computation over here. So, this is how this particular Adeline network works.