SimpleNN: a Simple MATLAB/Octave Package for Convolutional Neural Networks (CNN)

Machine Learning Group at National Taiwan University

Introduction

SimpleNN is a simple MATLAB/Octave package used for training CNN. Currently, this package supports two optimization methods:
  1. Newton method.
  2. Stochastic gradient method (SG) with learning rate decay and momentum.
The implementation is intuitive and well-optimized. Please see the following paper for more details:

C.-C. Wang, K.L. Tan, C.-J. Lin. Newton Methods for Convolutional Neural Networks (supplementary materials).

If you find this tool useful, please cite the above work.

Download MATLAB/Octave Scripts

Please check this github directory and see instructions there for the practical use.

Code for Experiments

The experimental code (for generating tables in the paper) is available here. Please read the README file in the directory for experimental setup.

Contributors

Chien-Chih Wang, Kent Loong Tan, and Chih-Jen Lin.

Please send comments and suggestions to Chih-Jen Lin.