01.11.2014 Views

MACHINE LEARNING TECHNIQUES - LASA

MACHINE LEARNING TECHNIQUES - LASA

MACHINE LEARNING TECHNIQUES - LASA

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

89<br />

J<br />

j<br />

( µ<br />

1,....,<br />

µ<br />

K) = ∑∑ φ( x ) −φ ( µ<br />

i)<br />

=<br />

K<br />

i= 1 x ∈C<br />

K<br />

∑∑<br />

j<br />

j<br />

( x ) φ( x )<br />

( )<br />

2<br />

l j<br />

j l<br />

( x ) φ( x ) ∑ φ( x ) φ( x )<br />

2 φ<br />

l i<br />

∈<br />

− +<br />

m<br />

x , x ∈C<br />

i j<br />

j l<br />

( ) ∑ k( x , x )<br />

( m )<br />

( m )<br />

i= 1 x ∈C i i<br />

⎛⎛<br />

K<br />

⎜⎜<br />

= ∑ ∑ ⎜⎜k x , x<br />

j i<br />

⎜⎜<br />

⎝⎝<br />

i<br />

i<br />

⎛⎛<br />

⎜⎜<br />

⎜⎜φ<br />

⎜⎜<br />

⎝⎝<br />

j j x C<br />

∑<br />

2 k x , x<br />

j i<br />

∈<br />

− +<br />

m<br />

i i x C<br />

∑<br />

j l i<br />

x , x ∈C<br />

i= 1 x ∈C i i<br />

j l i<br />

2<br />

⎞⎞<br />

⎟⎟<br />

⎟⎟<br />

⎟⎟<br />

⎠⎠<br />

2<br />

⎞⎞<br />

⎟⎟<br />

⎟⎟<br />

⎟⎟<br />

⎠⎠<br />

(5.26)<br />

In kernel K-Means, the iteration stops when the change in J is sufficiently small.<br />

As for all kernel methods, a clear bottleneck of kernel K-Means is that each iteration steps<br />

2<br />

requires O( M)<br />

computation step. Smilarly to classical K-means, kernel K-Means is also very<br />

sensitive to the initialization of the centers and may lead to poor partiotioning with a poor<br />

initialization, as shown in Figure 5-4.<br />

Figure 5-4: A difficult clustering problem. Standard K-Means (left) is unable to correctly separate the three<br />

clusters. Using a Gaussian kernel (gamma = 0.01) Kernel K-Means (right) successfully separates the three<br />

clusters. [DEMOS\CLUSTERING\KERNEL-KMEANS.ML]<br />

© A.G.Billard 2004 – Last Update March 2011

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!