Local Clustering Algorithm

1
Local Clustering Algorithm

• DISCOVIR
• 9-16-2002

2
Current Situation

• Image collection within a client is modeled as a
  single cluster.

3
Proposed Improvement

• Multiple clusters exist in the image collection

4
Group of similar local cluster A
5
Group of similar local cluster B
6
Group of similar local cluster C
7
Clustering Algorithm

• 3 clustering algorithms are proposed and tested
• C set of cluster center
• X dataset
• Goal of clustering minimize Error

8
Procedure

• Randomly pick k xj from X and assign them as
  the set C as initial cluster center.

Input datasetand cluster
9
Shifting Mean (SM)

• Suppose xj is picked and ci is the closest
  cluster center
• Let p be number of times ci wins, initially p1
• Update ci by

10
Competitive Learning (CL)

• Update ci by
• t the current number of iteration so far
• T total number of iteration intend to run
• We choose ? by 0.5, 0.3, 0.1

11
Illustration
ci
xj
12
Illustration
ci
xj
Winner (move closer)
13
Rival Penalized Competitive Learning (RPCL)

• Suppose cl is the second closest cluster center
  to xj
• Update ci by
• Update cl by
• We choose ? 0.05?

14
Illustration
ci
xj
unchanged
Rival (move away)
Winner (move closer)
15
Final Steps

• For each xj ,find the closest ci and mark xj
  belongs to ci
• Calculate error function
• Carryout experiments by varying of iteration,
  learning rate

16
Results
0.3 Error (400) Error (1000) Error (5000)
SM 1511 1982 2277
CL 1435 1798 1899
RPCL 1510 1974 2023
Fixed Learning rate Varying iteration
0.1 Error (400) Error (1000) Error (5000)
SM 2312 2008 1861
CL 2298 1888 1609
RPCL 2414 1948 1607
Fixed Learning rate Varying iteration
5000 Error (0.5) Error (0.3) Error (0.1)
SM 2290 2277 1861
CL 1627 1899 1609
RPCL 2248 2023 1607
Fixed IterationVarying learning rate
17
Screen Capture
18
Screen Capture
19
Screen Capture
20
Others

• Other variation
• ?i initial learning rate
• ?f final learning rate
• Interesting link for competitive learningsome
  competitive learning methods