Machine Learning in Excel
With just 4 neurons we can build a machine learning tool in an Excel spreadsheet !
(Sorry, In the video I kept saying it was 2 neurons!)
(Also I mean Sine! Not Sin, that's a completely different matter!)
At a basic level, a neural network is comprised of five main components: input(s), weights, bias, an activation function and an output.
The neural network in the video uses just 4 neurons & two 'Hidden Layers'. Between them, and after just a handful of times around the learning loop, it learns how to predict the Sine of an angle! The Feed-forward calculations are straightforward, however, the backpropagation (differential calculus using the calculus chain-rule) can be a different story! Fortunately, excel has in in-built 'solver' and it is this that hunts for and finds a (not THE) 'minimum error' to self correct the answers and learn. Feedforward and back-propagation techniques have been used with great effect in this example.
The time it takes to learn can be tedious, but it learns with a very few data points (in this case just 5) and provides an excellent (& accurate) answer.
...