Japanese / English

Detail of Publication

Text Language Japanese
Authors Masashi Tada,Tomoyuki Muto,Masakazu Iwamura,Koichi Kise
Title Extensions of Approximate Nearest Neighbor Search Based on a Multi-Valued Expression of Closeness to General Distributions
Journal Proc. First Forum on Data Engineering and Information Management
Presentation number F3-2
Month & Year February 2010
Abstract Approximate nearest neighbor search is a technique which greatly reduces processing time and required amount of memory for nearest neighbor search. We proposed two approximate nearest neighbor methods: MVH1 and MVH2. We comfirmed the effectiveness of these methods for uniform distribution with L1 norm. In this report, we get rid of the restrictions of these methods and propose efficient methods for Lp norm and general distributions. For L2 norm, improved MVH1 achived about 75% of processing time of LSH. Improved MVH2 achived about 50% of required amount of memory of LSH.
Back to list