OPEN HOUSE on Multi-Task and Complex Outputs Learning

Inductive transfer via embeddings into a common feature space

author: Shai Ben-David, University of Waterlo

Description

We consider the situation in which there is a basic learning task but different sub-tasks define different data generating distributions. Examples include learning to identify spam for various different email users, or parts-of-speech tagging for different text corpora. Our goal is to allow the use of training data coming from one sub-task for prediction under another sub-task distribution.

You might be experiencing some problems with Your Video player.

*Slides*
0:01	Inductive Transfer via Embedding into a Common Feature Space
0:28	Background
1:45	Inductive transfer for POS tagging
3:20	Structural Correspondence Learning (Blitzer, McDonald, Pereira)
9:26	Another potential application
10:23	Why does it work?
12:49	A formal framework
15:32	Our main Inductive Transfer Tool
16:22	The Common-Feature-Space Idea
16:41	Some more notation
18:31	Requirements from such embedding
19:34	Requirements from such embedding
20:53	How should one measure similarity of distributions?
22:55	A new distribution-similarity measure
25:33	Estimating dA from samples
28:38	A generalization bound
31:18	An algorithmic conclusion
35:07	Learning good embeddings
37:45	The resulting algorithm
40:03	A generalization bound for semi-supervised approach
42:01	Some experimental results
44:25	Comparison to a random projection
46:52	On the relationship between Inductive Transfer and Multi-Task Learning
48:07	Some other attempts of mine to ‘theoreize” practical heuristics
54:20	A generalization bound

Lecture rating

People found this lecture:
Worth seeing
because it is:
Valuable and informative
Well presented
Easily understandable
Acceptably recorded
You need to login to cast your vote.

Report a problem or upload files

If you have found a problem with this lecture or would like to send us extra material, articles, exercises, etc., please use our ticket system to describe your request and upload the data.
Enter your e-mail into the 'Cc' field, and we will keep you updated with your request's status.