A Study of Query-Based Dimensionality Reduction.
Augustine Shey Nsang, Anca Ralescu · 2011
This paper considers two approaches to query-based dimensionality reduction. Given a data set, D, and a query, Q, the first approach performs a random projec-tion on the dimensions of D that are not in Q to obtain the data set DR. A new data set (DRQ) is then formed comprising all the dimensions ofD that are in the query Q together with the dimensions of DR. The resulting data set (Q(DRQ) is obtained by applying the query Q to DRQ. A similar approach is taken in the second ap-proach with the difference that the random projection method is replaced by Principal Component Analysis. Comparisons are made between these two approaches with respect to the inter-point distance preservation and computational complexity.