Understand how a neural network learns, step by step

6/29/2019 - Category: ai, coding, research,

Last time, we trained a simple perceptron and saw how it learned over time. Today, let's train a neural network!

You will be able to clearly see and understand the learning steps by hitting the "Step By Step Training" button below.

Alternatively, you can hit the "Start Auto Training" button to trigger 10,000 learning cycles. When the training is done, you'll see that the neural network makes fewer and fewer errors over time, thus guessing more correctly. It's learning!

The method used is 'supervised learning.' This means we have a dataset with inputs and known answers (outputs) to train the neural network to correctly guess the answers over time. We'll use the simple XOR dataset.

I coded the following animation using JavaScript and p5.js on top of the awesome Daniel Shiffman's neural network toy library.

Grab the whole code on my GitHub page if you like!