Brian Cheung, PhD Student, UC Berkeley - Deep Learning Summit 2015

This presentation took place at the Deep Learning Summit in San Francisco on 29-30 January 2015. Supervised learning algorithms attempt to learn task relevant factors while being invariant to all others. In contrast, unsupervised learning algorithms seek latent factors which are relevant for a wide range of high level tasks. In this work, we combine these two ideas by augmenting autoencoders with a supervised learning cost to create a semi-super