NIPS Workshop on Kernel Learning: Automatic Selection of Optimal Kernels, Whistler 2008

The Sample Complexity of Learning the Kernel

author: Shai Ben-David, David R. Cheriton School of Computer Science, University of Waterloo
published: Dec. 20, 2008, recorded: December 2008, views: 325

Slides

Slides
0:00	The sample Complexity of Learning the Kernel
0:32	The modeling of prior knowledge
1:39	What happens in real-life?
3:12	“Automated selection of kernels” (1)
4:15	“Automated selection of kernels” (2)
5:26	Raising awareness to the sample costs of kernel learning
7:54	The sample complexity question
9:12	The “richness” of a family of kernels
10:00	A combinatorial measure of that richness
11:03	Bounds for specific kernel families
11:38	Other types of data for learning kernels
13:23	Prior-Knowledge Expression – A major (under researched?) Challenge.
13:31	Bounds for specific kernel families
14:05	Prior-Knowledge Expression – A major (under researched?) Challenge.
14:10	Other types of data for learning kernels
14:30	Prior-Knowledge Expression – A major (under researched?) Challenge.
15:56	Expressing SSL prior beliefs
17:32	An SSL Kernel learning challenge
18:21	The multi-task learning setting
19:33	Kernel learning in MTL
21:12	Additional task-relatedness assumptions.
23:11	Conclusions
24:42	- questions

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.

Lecture popularity: You need to login to cast your vote.

Description

The success of kernel based learning algorithms depends upon the suitability of the kernel to the learning task. Ideally, the choice of a kernel should based on prior information of the learner about the task at hand. However, in practice, kernel parameters are being tuned based on available training data. I will discuss the sample complexity overhead associated with such ”learning the kernel” scenarios. I will address the setting in which the training data for the kernel selection is target labeled examples, as well as settings in which this training is based on diﬀerent types of data, such as unlabeled examples and examples labeled by a diﬀerent (but related) tasks. Part of this work is joint with Nati Srebro.