/Neural-Squarer

JAX-based neural network for squaring numbers

Primary LanguagePython

Neural Squarer

A simple dense neural network written in python using JAX that approximates x^2

Building

run main.py in Linux or WSL2 (JAX does not have windows support)

How it works

The program is a dense neural network made of 5 densely connected 25-parameter layers that work to approximate the square of values -10 through 10. Type 'exit' to stop training.

Accuracy

After a few minutes on my RTX 3080, the accuracy of estimation is within 0.012 of the real value.