en-de
en-es
en-fr
en-sl
en
en-zh
0.25
0.5
0.75
1.25
1.5
1.75
2
Inductive transfer via embeddings into a common feature space
Published on Feb 25, 20073356 Views
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 diffe
Chapter list
Inductive Transfer via Embedding into a Common Feature Space00:01
Background00:28
Inductive transfer for POS tagging01:45
Structural Correspondence Learning (Blitzer, McDonald, Pereira) 03:20
Another potential application09:26
Why does it work?10:23
A formal framework12:49
Our main Inductive Transfer Tool15:32
The Common-Feature-Space Idea16:22
Some more notation16:41
Requirements from such embedding18:31
Requirements from such embedding19:34
How should one measure similarity of distributions?20:53
A new distribution-similarity measure 22:55
Estimating dA from samples25:33
A generalization bound28:38
An algorithmic conclusion31:18
Learning good embeddings35:07
The resulting algorithm37:45
A generalization bound for semi-supervised approach40:03
Some experimental results42:01
Comparison to a random projection44:25
On the relationship between Inductive Transfer and Multi-Task Learning46:52
Some other attempts of mine to ‘theoreize” practical heuristics48:07
A generalization bound54:20