IPAM Tutorials on Theano/Torch
On Deep Learning, Feature Learning July 9 - 27, 2012
Welcome to the Practical Sessions for the summer school
10 mins crash course in Python, numpy
10 mins crash course in Lua, Torch7
Remaining time - getting people into groups and setting them up to run the sample code on laptop or EC2. Once they get it running, they can go for lunch or stick around and play with things.
Models: SVM, MLP, ConvNets, (Logistic Regression?)
Data Sets: MNIST, CIFAR, Google Street View House Numbers (SVHN). SVHN is an interesting new data set, very few results are available at this time (and is more computer visionny that MNIST).
Optimization Methods: SGD, ASGD, L-BFGS; batch vs. mini-batch vs. online
Python: Imprinting, K-Means, Autoencoder, De-noising Autoencoder, RBM, (Sparse Coding?)
Torch: Linar Autencoder, Convolutional Autoencoder, Linear and Convolutional PSD (Predictive Sparse Decomposition) Autoencoder
Persitant Contrastive Divergence?
Recurrent Neural Networks?
GPU Programming 101?
Torch/nn extensions: write your own modules