Similarity Measures for Nearest Neighbor Search and Classification Methods in High Dimensional and Large Number of Data

2016 Fiscal Year Final Research Report

• Project/Area Number: 25730142
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Intelligent informatics
• Research Institution: Yamagata University (2015-2016); National Institute of Genetics (2013-2014)
• Principal Investigator: Suzuki Ikumi 山形大学, 大学院理工学研究科, 助教 (20637730)
• Project Period (FY): 2013-04-01 – 2016-03-31
• Keywords: ハブネス / ハブの軽減 / センタリング / 近傍法

Outline of Final Research Achievements

Recently, hubness, a phenomenon occurring in high-dimensional datasets as a result of curse of dimensionality has attracted the attention of researchers in the artificial intelligence community, especially for data mining and machine learning.
In this work, we pointed out that the hubness influences the performance of k-nearest neighbor (k-NN) methods. We reported that subtracting mean vector from each sample (centering) is a simple, yet very effective for improving k-NN classification. Also, we proved that centering is effective for k-NNs, because centering reduces hubs in a dataset.

Free Research Field

統計的機械学習