event thumbnail image
Sessions

A Simpler Unified Analysis of Budget Perceptrons

author: Ilya Sutskever, Department of Computer Science, University of Toronto

Description

The kernel Perceptron is an appealing online learning algorithm that has a drawback: whenever it makes an error it must increase its support set, which slows training and testing if the number of errors is large. The Forgetron and the Randomized Budget Perceptron algorithms overcome this problem by restricting the number of support vectors the Perceptron is allowed to have. These algorithms have regret bounds whose proofs are dissimilar. In this paper we propose a unified analysis of both of these algorithms by observing that the way in which they remove support vectors can be seen as types of $L_2$-regularization. By casting these algorithms as instances of online convex optimization problems and applying a variant of Zinkevich's theorem for noisy and incorrect gradient, we can bound the regret of these algorithms more easily than before. Our bounds are similar to the existing ones, but the proofs are less technical.

You might be experiencing some problems with Your Video player.
Slides
0:00 A simpler unified analysis of budget perceptrons
0:14 Summary and Conclusion
0:50 Perceptrons (1)
1:26 Perceptrons (2)
1:37 Kernel Perceptrons
2:13 Kernel Perceptron (1)
2:30 Kernel Perceptrons (2)
3:02 Kernel Perceptrons (3)
3:30 The Budget idea
4:41 The (simplified) Forgetron (1)
4:59 The (simplified) Forgetron (2)
6:10 The Randomized Budget Perceptron (1)
6:30 The Randomized Budget Perceptron (2)
6:42 Regret Bounds
7:46 The main insight (this part is new)
9:32 Specifically: The (simplified) Forgetron
10:32 Specifically: The Randomized Budget perceptron
11:24 Justification
12:13 Online convex programming (1)
12:45 Online convex programming (2)
13:28 icml09_sutskever_asua_01_Page_28
14:13 The Forgetron
15:44 The Randomized Budget Perceptron
16:22 Bonus: tracking
17:05 Thank you!
17:11 - Questions

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.

Link this page

Would you like to put a link to this lecture on your homepage?
Go ahead! Copy the HTML snippet !

Write your own review or comment:

make sure you have javascript enabled or clear this field: