/training-neural-nets

Training a deep neural network

Primary LanguageMATLAB

Neural Nets

In this project I accomplished two goals:

Using a Two Layer Neural Net as a function approximator

Without using tools from MATLAB's neural net toolbox, I created a two-layer neural net that computes a piecewise linear approximation of an input function f.

  • layer 1 uses the ReLU nonlinearity
  • layer 2 does not use nonlinearity since it uses the ReLU as the identity

Example

Training a Small Neural Net

A small neural net was trained on data with binary labels. Network structure and training parameters were tuned until achieving 97.12% testing accuracy.

Test Data

Classification Results (97.12% accuracy)