Can a Neural Net Learn It?

A neural network is a type of function originally inspired by how the brain works. However, what makes it particularly useful in machine learning are the following properties:

1. Neural networks can approximate any continuous function (universal approximation theorem)
2. We have an effective way to fit a neural network to input data, called backpropagation.
3. Using the neural network (forward propagation) and training the network (backpropagation) can both be done very efficiently using existing hardware (GPUs).

Below you can experiment with fitting a small neural net to a function and watch as it learns. You can select sample functions from the dropdown, write your own custom function, and specify the (fully connected) neural network structure.