Title: Multi-label Learning - Part II
Speaker: Purushottam Kar
Day: October 8, 2015, Thursday
Time: 5:15 PM
Venue: KD101 (CSE main)

Abstract:
We will continue our discussion on multi-label learning. In the interest of those who could not attend the last meeting given the rescheduling, we will recapitulate the previous talk material as well.

To refresh one's memory, multi-label learning allows us to answer questions of the sort: Given an image, can I learn how to tag it with all the objects present in that image? Given a document, can I tag it with all the topics the document talks about? This has several applications in document tagging, recommendation systems (e.g. assigning a NetFlix user movies (s)he is likely to watch), and ranking tasks.

We shall explore three solution strategies for this problem: binary relevance, embedding methods, and tree methods.