e-nuts.net

Feed forward neural networks

A neural network is a computer paradigm, named after human brain's structure. On the internet you can read a lot of documents about neural networks. In order to make concepts less complex, you can imagine that a feed forward neural network with supervised learning works like a simple function estimator. If you train it with some examples of input/output, your network will returns right outputs also for inputs you've never used during training.

Grouper

Considering that all neural networks work like a black-box, the best way to test if they work is to write some real applications. So I created "Grouper" that is a simple application that use a small neural network (made up of about 20 neurons). User can draw some points of two colours on screen. This app shows how neural network understand colour allotment subdividing screen on different colour area.

A simple evolution movie

Here you can watch two movies showing how Grouper evolves during time. A light colour means a better confidence of network about results. This movie is slowed down: on real-time this jobs is completed by a common machine in a couple of seconds. In first movie there's a fixed set of points. In the second one points are added dynamically during training