/UFLDL-Tutorial

Implementations of some common and simple deep learning algorithms

Primary LanguageMATLABMIT LicenseMIT

UFLDL-Tutorial

This repository contains the MATLAB implementation of the exersises from the Unsupervised Feature Learning and Deep Learning (UFLDL) Tutorial.

Notes about the input data

You can find the appropriate data for each exercise at:

  1. For the sparseae_exercise, you do not need to do anythings as the data are already included (it is the IMAGES.mat file).
  2. For the sparseae_exercise_vector, you do not need to do anything as the data are already included (it is the IMAGES.mat file).
  3. For the pca_2d, you do not need to do anything as the data are already included (it is the pcaData.txt file).
  4. For the pca_exercise, you do not need to do anything as the data are already included (it is the IMAGES_RAW.mat file).
  5. For the softmax_exercise, you need to download the MNIST dataset. Specifically, you need to download: * The train-images-idx3-ubyte.gz * The train-labels-idx1-ubyte.gz * The t10k-images-idx3-ubyte.gz * The t10k-labels-idx1-ubyte.gz More information can be found on the tutorial's webpage.
  6. For the stl_exercise, you need to download the same dataset as in the softmax_exercise (see 5).

A note about the license

The initial code for the excercises is available at the UFLDL Tutorial, and my code is provided here under the MIT licence. But, as noted in the Sparse Autoencoder exercise, xthe minFunc subdirectory is a 3rd party software that implements the L-BFGS optimization algorithm and is licensed under a Creative Commons, Attribute, Non-Commercial licence. If you need to use this software for commercial purposes, you can download and use a different functions (fminlbfgs) that can serve the same purpose, but runs ~3x slower. You can find more about this function here.