Title :
Fast Density Estimation for Approximated k Nearest Neighbor Classification
Author :
Kobayashi, Takao ; Shimizu, Ikuko
Author_Institution :
Dept. of Comput. & Inf. Sci., Tokyo Univ. of Agric. & Technol., Koganei, Japan
Abstract :
We propose a method for fast density estimation of samples, which makes it possible to significantly accelerate classification based on the k nearest neighbor (kNN) method. Our main premise is that many trials of a rough estimation of probability density function are conducted, and they are integrated by Bayes´ theorem. The experimental results indicated that the classification time used in our method was at least 30 times faster than that of kNN.
Keywords :
approximation theory; pattern classification; approximated k nearest neighbor classification; fast density estimation; probability density function; bayes theorem; k nearest neighbor method; locality sensitive hashing; partition of a space;
Conference_Titel :
Frontiers in Handwriting Recognition (ICFHR), 2010 International Conference on
Conference_Location :
Kolkata
Print_ISBN :
978-1-4244-8353-2
DOI :
10.1109/ICFHR.2010.60
