event thumbnail image
Machine Learning Summer School 2006 - Taipei
Pascal

The stability of a good clustering

author: Marina Meila, Department of Statistics, University of Washington

Description

If we have found a "good" clustering C of data set X, can we prove that C is not far from the (unknown) best clustering C* of this data set? Perhaps surprisingly, the answer to this question is sometimes yes. We can show bounds on the distance( C, C* ) for two clustering cost functions: the Normalized Cut and the squared distance cost of K-means clustering. These bounds exist in the case when the data X admits a "good" clustering for the given cost.

You might be experiencing some problems with Your Video player.
Slides
0:05 The Stability of a Good Clustering
1:19 Optimizing these criteria is NP-hard
4:01 Results summary
5:25 A graphical view
5:56 A graphical view 01
6:09 Overview
6:43 The Confusion Matrix
7:28 The Misclassification Error distance
8:52 Representing clusterings as matrices
10:27 Representing clusterings as subspaces
11:09 Why K-1 dimensions? Intuition
13:28 Distortion is quadratic in X
14:27 Quadratic functions and the eigengap
14:55 Quadratic functions and the eigengap 01
17:10 Solve relaxed minimization
20:42 Main Theorem
21:43 Bound for d(X,Xopt)
23:21 Extensions
23:42 The normalized cut (Shi & Malik 99)
24:54 An improved bound for NCut
26:36 Experiments – bounds for best clustering
27:55 Experiments – bounds for best clustering 01
29:07 Experiments – random perturbation of Xopt
30:08 Conclusions

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: