(T) The Silicon Valley IEEE AI group had an interesting lecture last month, that I attended, from Stanford’s Professor Stefano Ermon. Many of the recent successes of machine learning have been characterized by the availability of large quantities of labeled data. Nonetheless, we observe that humans are often able to learn with very few labeled examples or with only high-level instructions for how a task should be performed.
In this talk, Professor Ermon presented some new approaches for learning useful models in contexts where labeled training data is scarce or not available at all, and in particular:
1) He introduced new techniques for learning generative models, including the use of random projections to simplify probabilistic models while preserving most of the information, and a new boosting framework to learn ensembles of models.
2) He discussed ways to use prior knowledge (such as physical laws) to provide supervision, showing how we can learn to solve useful tasks, including object tracking, without any labeled data.
3) Finally, he introduced new approaches to leverage spatio-temporal structure in semi-supervised learning frameworks and present applications of these ideas to address development and sustainability issues, including new scalable methods to map poverty and monitor food security in developing countries using satellite imagery.
Following is the video of Professor Ermon’s talk:
And his slide presentation: Learning_Ltd_Supervision_2017_11_01
Note: The picture above is a Child’s toy.
Copyright © 2005-2017 by Serge-Paul Carrasco. All rights reserved.
Contact Us: asvinsider at gmail dot com.