Making QBIC Faster (54)

Clustering

Reducing to k candidates is no better than kN_O candidates if each one comes from a separate page (recall than N_O may be larger than 100)

This situation becomes worse when vectors are smaller, and so N_O becomes larger

Approach: try to ensure that "similar" objects are placed in the same database page.

Plan to look at how effectively clustering can be performed for very large numbers of dimensions.