617-384-9091
iacs-info@seas.harvard.edu
Wednesday, January 13, 2016
1:30pm - 4:30pm
Classical machine learning techniques often involve extracting some form of manually designed features from data and then training a model for classification. Designing the features is an important task, often involving domain knowledge and manual tuning. Deep learning methods instead learn multiple levels of feature representations directly from the data. Learning the features has been shown to improve classification results, and make the same model applicable to different data modalities like images, speech, and text. Theano is a great Python library, that facilitates using deep learning methods, even on the GPU. In this workshop we will introduce the basics of deep learning and models like deep networks, autoencoders, and convolutional networks, and how to train them.
Part one (Wednesday, January 13) will focus on deep learning fundamentals and how to learn features in a supervised and unsupervised setting, as well as best practices for training these complex models.
Prerequisites: Python programming, Laptop with Python 2.7, IPython, and Theano installed for in-class work.
[NOTE: Part two of this workshop is offered on Thursday, January 14 at 1:30pm and will introduce convolutional neural networks and cover advanced tips and tricks for training. Register separately for Thursday's workshop at http://computefest.seas.harvard.edu/workshops]