In the last section, you learned about a single artificial neuron and used it to predict the energy output. If we compare it with the linear regression result of Chapter 3, Machine Learning for IoT, we can see that though the single neuron did a good job, it was not as good as linear regression. The single neuron architecture had an MSE value of 0.078 on the validation dataset as compared 0.01 of linear regression. Can we make it better, with maybe more epochs, or different learning rate, or perhaps more single neurons. Unfortunately not, single neurons can solve only linearly separable problems, for example, they can provide a solution only if there exists a straight line separating the classes/decision.
The network with a single layer of neurons is called simple perceptron. The perceptron model was given by Rosenblatt...
