Assignment 4

Unsupervised Feature Learning and Deep Learning

Submission: December 3rd

Maximum of 3 students per group

• 1. Follow the tutorial in http://ufldl.stanford.edu/wiki/index.php/UFLDL_Tutorial. The tutorial has to be entirely implemented in Python and it has to be presented as a unique IPython notebook. The exercises of the tutorial that have to be implemented (each one must corresponde to a section in the notebook) are:

     • Sparse Autoencoder

     • Vectorization

     • PCA in 2D

     • PCA and Whitening

     • Softmax Regression

  2. Use a dataset of face images instead of the handwritten digits dataset suggested in the tutorial. You can use any of the datasets in http://www.face-rec.org/databases/.

  3. The notebook must clearly present the code, explain it, present the results and discuss them.

  4. The code has to be develop by the students, plagiarism will invalidate the assignment.