Data Clustering a very short introduction

1
Data Clustering(a very short introduction)
Intuition grouping of data into clusters so that
elements from the same cluster are more similar
to each other than they are to a element from a
different cluster.
2

• Major decision steps
• Pattern representation
• Definition of a pattern proximity measure
• Method for clustering
• Data abstraction (cluster representation, e.g.
  centroids )
• Assessment of output (evaluation of the output)

3
Proper pattern representation can simplify
clustering
Identify circles Cartesian coordinates? -gt use
polar coordinates (r,?)
4

• Definition of a pattern proximity measure
• Similarity between two clusters C1 xi C2 yj
• Examples
• Single Link minij dist(xi,yj) lt e
• Complete Link maxij dist(xi,yj) lt e

5
Pose Clustering

• Aims to solve the LCP problem.
• Compute a set of transformations that align one
  structure with the other.
• Cluster transformations.
• Check large clusters.
• Idea a large common point set will produce a
  large number of similar transformations.

6
Example Clustering of 3D transformations Goal
Prevent redundant solutions Representation 1)
3x3 matrix 1x3 vector Problem how to
measure distance between two
transformations? 2) Image of points, T(S)
dist(T1,T2) dist(T1(S),T2(S)) (for
example RMSD or bottleneck) Problem time
complexity Solution use less points, 3-4
farthest points are stable enough