spectral clustering
IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS, VOL. 24. NO. 9, SEPTEMBER 1994
1417
. . .
:;.
1 Fig. 1. A density function f on a set[ a, b] .
. . .
. ... . .. .:'. ........... '... .... . .: . . .. .. .. . . . . . . ., . .:, .. ')
. ., . . ..
e .
. . . . . . .... . -' . :E..
,;
I,,
,. .
. .. . .
"
I
. . .: . ..
"
,
.I
.
Count the number of points in S, n (say). Let&=number of representative points
. . . . . . . . .'. . .:,,:,,.. ..: .. . . . .. ... .. . ... ... . . . . . . .. . . . . . . . .. . . . . . .. . . . . . . . '
. ._.
a:.
,
.
Y.,
'
,
"
required
:
.:
. .
. . . . . . . . . 4 ... . . . .: . .
.
1 Find h n
take 8: h, I
-
= Compute ml, m2, and find I$,, in,*,
- - -mn - . . m*,
Fig. 3. A triangular distribution data of size 200. (a) Local best representative point. (b) Cluster center by applying C-means algorithm (C= 1). (c) Almost 5% representative points. (d) Almost 10% representative points. representative points may be considered as the generalization of mode estimation and seed point detection problem. In his classical work[l], Parzen showed how to construct a family of consistent and asymptotically unbiased estimates of probability density function and its mode. The work
was extended by Cacoullos[ 2], among others[9]. In the literature of cluster analysis, there exist several approaches of seed point estimation[8]. Macqueen 131 chooses the first k data units in data set as the initial seed points. Forgy[4] takes any desired partition of the dta units into k mutually exclusive groups and computes the group centroids as seed points. Astrahan[5] computes the density for each data unit as the number of other data units within some specified distance, orders the data units by density, and chooses the one with the highest density as the first seed point. The subsequent seed points are chosen in order of decreasing density, subject to the stipulation that each new seed point is at least a minimum distance away from all other previously chosen seed points. A simpler approach was suggested by Ball and Hall[6]. Here, the overall mean vector of data set is considered as the first seed point. The subsequent seed points are selected by examining the data units in their input sequence and accepting any data unit which is at least some specified distance, say d, from all previously chosen seed points. The process is continued until k seed points are accumulated or the data set is exhausted. Ling[ 111 suggests that each (k,r)-cluster has the property that its elements are within a distance T of at least k other elements of the same cluster, and the entire set can be marked by a chain of links each of length less than or equal to r. But there are no guidelines about k and r. Given n data, the problem is that of finding a small subset of no<< n data that provides good representation of the original data. Let a representative point p represent r data in its neighborhood. Intuitively, p should be such that the sum of its distances from these r data is minimum. But an exact algorithm, which gives the optimal representative point set, is computationally very expensive. See Section VI for details.
find k Find the sets 'SI and find
1=1,2,.-. - - s t
Decrease
I v*
t-l
J
yes
8
Fig. 2. A flowchart of the proposed algorithm AL-2.

each data object is in exactly one subset ? ... – A cluster is a set of points such that ...“representative” point of a cluster 4 center-...
Analysis is the process of finding a ? subset D’ of D [D: data set to be analyzed...Sampling (determines a representative subset of the...
FewSparse Modeling for Finding Representative Objects_文学研究_人文社科_专业资料...few representatives for a dataset, i.e., a subset of data points that ...
Finding groups of objects such that the objects ...“representative” point of a cluster 4 center-... which is a subset of the points of the ...
each data object is in exactly one subset ? ...A cluster is a set of points such that any ...“representative” point of a cluster 4 center-...
each data object is in exactly one subset ? ... – A cluster is a set of points such that ...“representative” point of a cluster 4 center-...
each data object is in exactly one subset ? ...A cluster is a set of points such that any ...“representative” point of a cluster 4 center-...
? Finding groups of objects such that the ...“representative” point of a cluster 4 center-...(a) Green points denote the data set in a ...
? Finding similarities between data according to ...K-Medoids Clustering: Find representative objects (...points in the test set and the closest centroids...
‘gender’ and a representative population of ...data points are summarized in a single figure ?...Selection may involve the chossing a subset of ...
热门文档