From 7a4a548c64aa73a7b212be2bbd194eaf176a8a5f Mon Sep 17 00:00:00 2001 From: rriegel Date: Fri, 8 Jun 2007 17:21:59 +0000 Subject: [PATCH] mention of trees --- fastlib/u/rriegel/nips07/gnp_nips07.tex | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) diff --git a/fastlib/u/rriegel/nips07/gnp_nips07.tex b/fastlib/u/rriegel/nips07/gnp_nips07.tex index 295a994c37..7441177158 100644 --- a/fastlib/u/rriegel/nips07/gnp_nips07.tex +++ b/fastlib/u/rriegel/nips07/gnp_nips07.tex @@ -930,13 +930,13 @@ storing summary results during the process of iterative refinement. Affinity propagation \cite{affinity} is a recent clustering technique that chooses {\em exemplars} from a data set $X \subset \mathcal{X}$ -such in attempt to maximize the sum of similarities between all points -and their nearest exemplar. Two points $x_i$ and $x_j$ have -similarity $\simil{i}{j}$, and special case $\simil{i}{i}$ is set to -parameter $p$, the {\em preference} of points to be exemplars. The -number of clusters to find is not explicitly specified, but is -positively correlated with $p$. A typical value of $p$ is the median -of the similarities between all pairs of points. +in attempt to maximize the sum of similarities between all points and +their nearest exemplar. Two points $x_i$ and $x_j$ have similarity +$\simil{i}{j}$, and special case $\simil{i}{i}$ is set to parameter +$p$, the {\em preference} of points to be exemplars. The number of +clusters to find is not explicitly specified, but is positively +correlated with $p$. A typical value of $p$ is the median of the +similarities between all pairs of points. The algorithm alternateingly updates messages $\respo{i}{j}$ and $\avail{i}{j}$ with