An artificial neural network, for short, is a machine learning algorithm based on a very crude approximation of the way we used to believe brains, the neurons in the human brain work or an animal brain work.
We now know that it's actually quite a bit more complicated than this, but the mathematical model is actually quite useful.
It works really well for solving a series of problems, so one starts with a biological neural network.
Each neuron has models, there's a set of inputs that are coming in to the neuron and a set of outputs leading to other neurons.
Each neuron receives an input value from its predecessor, neurons, these connections coming in and then the values are essentially all summed in the first half of the neuron.
The neuron also contains something called an activation function which changes the behavior of the output. So, given a certain input, how much should we be outputting on that neuron.
Each of these connections coming into the neuron contain something called a weight, and the weight can be adjusted in order to change how the data is coming into the.
How much information is coming in or how much or how little information is coming in there's also an additional weight here called a bias.
This bias essentially just shifts this activation function, either to the left or to the right once again changing the behavior, how it's firing based upon the input that's coming into it.
So each of these weights is essentially like a knob that can be tuned to change the behavior of the neuron.
So, by tuning all these knobs, essentially we we're changing the overall way data is flowing through this neural network, and I should probably point out that it's essentially just numbers coming in to each of these values in numbers coming out of each of these neurons.
Then one has to wire up this network of neurons into a graph containing edges and nodes.
These artificial neural networks are composed of an input layer which you can see over here and an output layer over here, so data flows into the input layer and comes out of the output layer.