Talk:Vector quantization: Difference between revisions

Revision as of 09:15, 7 June 2011

Revision as of 15:29, 31 January 2013

Line 1:

==Some math==

During some of my colleges I got some math what could be nice to be on this page. only I don't have enough mathimatical background to prove the used maths.

= The Math =

create set of prototypes = <math>w^1,w^2,\dots,w^K, w^k \in \mathbb{R}^n</math>

the data = <math>\xi^1,\xi^2,\dots,\xi^p, \xi^k \in \mathbb{R}^n</math>

by using the [[Euclidean_distance#Squared_Euclidean_Distance|Squared_Euclidean_Distance]]

we can determine the multidimention distance between a prototype and a data point.

<math>L_2(a,b) = \sum_{i=1}^d|a_i-b_i|</math>

Based on this we can find the closest prototype to a given datapoint.

<math>w^{i*} = argmin_j \{d[w^j,\xi]\}</math>

assign <math>\xi</math> to prototype <math>i^*</math>

This way the winner takes it all and the closest prototype should be moved using:

<math> w^{i^*} \xi \to w^{i^*} + \eta( \xi^t - w^{i^*})</math>

where <math>\eta < 1</math> is the learning rate

[[User:Spidfire|Spidfire]] ([[User talk:Spidfire|talk]]) 15:29, 31 January 2013 (UTC)

==Untitled==

also want to see pictures <span style="font-size: smaller;" class="autosigned">—Preceding [[Wikipedia:Signatures|unsigned]] comment added by [[Special:Contributions/138.246.7.74|138.246.7.74]] ([[User talk:138.246.7.74|talk]]) 13:50, 15 July 2010 (UTC)</span><!-- Template:UnsignedIP --> <!--Autosigned by SineBot-->