Illustration of the LGB Vector Quantization Algorithm

This presentation has been prepared with webslider. The slides may be perused by selecting any of the entries in this Table of Contents.
  1. Training vectors as points in a 2D space
  2. First step, the centroid of the entire set
  3. Split the centroid into two (randomly)
  4. Nearest-neighbor search, result in 2 clusters
  5. Update the centroid from each cluster
  6. Repeat nearest-neighbor search
  7. And repeat centroid update
  8. Until the centroids are stable (end of the inner loop)
  9. Repeat outer loop, double the number of centroids by splitting
  10. Iterate inner loop until the centroids are stable
  11. Continue...
  12. 8 centroids
  13. And...
  14. Finally, stop when the desired number of centroids is obtained

Copyright © Minh N. Do 2000