Past Recording
Share
Star()
Learning deep representations by mutual information estimation and maximization
Thursday Apr 11 2019 22:30 GMT
Please to join the live chat.
Learning deep representations by mutual information estimation and maximization
Why This Is Interesting

In this work, we perform unsupervised learning of representations by maximizing mutual information between an input and the output of a deep neural network encoder. Importantly, we show that structure matters: incorporating knowledge about locality of the input to the objective can greatly influence a representation’s suitability for downstream tasks. We further control characteristics of the representation by matching to a prior distribution adversarially. Our method, which we call Deep InfoMax (DIM), outperforms a number of popular unsupervised learning methods and competes with fully-supervised learning on several classification tasks. DIM opens new avenues for unsupervised learning of representations and is an important step towards flexible formulations of representation-learning objectives for specific end-goals.

Time of Recording: Thursday Apr 11 2019 22:30 GMT
Slides: please to see content