A network of "neurons" on different layers, where each neuron is a function of the neurons in the previous layer.
The input is any set of numbers, and the output is usually a classification, the percentage chance that is each of various categories.
The way the coefficients of a function are set is by training the neural network with training
data, basically questions you
know the answers to. Uses a
loss function to determine how far off the answer the AI comes up with are from the real answer, and using gradient descent (goes down the path of the function with the steepest partial derivative) to find the minimum value of the
loss function (most
correct answers) to find the coefficients of the function that lead to the highest accuracy