Neural Networks - A (very*) light introduction
This project illustrates how a simple feedforward neural network learns parameters.
The visualization is an introduction to Neural Networks with a view to facilitate the learning process for incoming Machine Learning students. With this intuition we expect that it will be easier to learn more complex concepts.
The animations and illustrations should provide intuition, but are not designed to be comprehensive. For this reason, we intentionally skip many details.
The expected audience are students with some very rudimentary understanding of linear models, and a notion of what a Neural Network is. A background in math is not required.
We introduce concepts with a mere 6-8 data points arbitrarily selected to illustrate a problem. We then use those data points to pre-train two simple models and capture intermediate data such as weights and losses, for the chart animations. This data lives in the project github repository.
Next we generate a real life example dataset of weights and losses by training a user defined neural network model on the fly using the MNIST dataset.