Clustering and the K-means algorithm - MIT...

24
Clustering and the K-means algorithm Yihui Saw 18.304 Seminar Talk I March 6, 2013 Saturday, March 16, 13

Transcript of Clustering and the K-means algorithm - MIT...

Clustering and the K-means algorithm

Yihui Saw18.304 Seminar Talk I

March 6, 2013

Saturday, March 16, 13

Clustering examples

• Customer purchase patterns

• Language family models

• Data compression

Saturday, March 16, 13

Original Image

Saturday, March 16, 13

2 colors

Saturday, March 16, 13

4 colors

Saturday, March 16, 13

8 colors

Saturday, March 16, 13

The clustering problem

Saturday, March 16, 13

Distance metric

Saturday, March 16, 13

The cost of clustering

Saturday, March 16, 13

K-means algorithm

Saturday, March 16, 13

K-means algorithm

Saturday, March 16, 13

K-means algorithm

Saturday, March 16, 13

K-means algorithm

Saturday, March 16, 13

K-means algorithm

Saturday, March 16, 13

K-means algorithm

Saturday, March 16, 13

Proof of convergence• Each iterative step necessarily lowers the

cost - the cost monotonically decrease

Saturday, March 16, 13

Proof of convergence• Each iterative step necessarily lowers the

cost - the cost monotonically decrease

]

Saturday, March 16, 13

Convergence to local minimum

Saturday, March 16, 13

Convergence to local minimum

Saturday, March 16, 13

Convergence to local minimum

Saturday, March 16, 13

Convergence to local minimum

Saturday, March 16, 13

Convergence to local minimum

Saturday, March 16, 13

Convergence to local minimum

Saturday, March 16, 13

Convergence to local minimum

Saturday, March 16, 13