This simple Neural Network tries to discriminate points in a plane.
It is a javascript implementation of a fully connected Neural Network. It uses the Sigmoid activation Function.
A Gradient-free "Optimization" method is used:
- One picks a random direction in the Parameter Space
- One follows this direction as long the Error goes down
- If the Error goes up again it starts at step 1
Watch Live training and Results here https://evanwonisch.github.io/random-walk-optimisation/
The documentation is beeing worked on.