The Influence of Hubness on NN-Descent

2019 ◽  
Vol 28 (06) ◽  
pp. 1960002 ◽  
Author(s):  
Brankica Bratić ◽  
Michael E. Houle ◽  
Vladimir Kurbalija ◽  
Vincent Oria ◽  
Miloš Radovanović
Keyword(s):  
Nearest Neighbor ◽  
Synthetic Data ◽  
Machine Learning Algorithms ◽  
Data Sets ◽  
Accurate Approximation ◽  
K Nearest Neighbor ◽  
Major Drawback ◽  
Neighbor Graph ◽  
Nearest Neighbor Graph ◽  
The Cost

The K-nearest neighbor graph (K-NNG) is a data structure used by many machine-learning algorithms. Naive computation of the K-NNG has quadratic time complexity, which in many cases is not efficient enough, producing the need for fast and accurate approximation algorithms. NN-Descent is one such algorithm that is highly efficient, but has a major drawback in that K-NNG approximations are accurate only on data of low intrinsic dimensionality. This paper represents an experimental analysis of this behavior, and investigates possible solutions. Experimental results show that there is a link between the performance of NN-Descent and the phenomenon of hubness, defined as the tendency of intrinsically high-dimensional data to contain hubs – points with large in-degrees in the K-NNG. First, we explain how the presence of the hubness phenomenon causes bad NN-Descent performance. In light of that, we propose four NN-Descent variants to alleviate the observed negative inuence of hubs. By evaluating the proposed approaches on several real and synthetic data sets, we conclude that our approaches are more accurate, but often at the cost of higher scan rates.

DRSA: a non-hierarchical clustering algorithm using k-NN graph and its application in vegetation classification

2015 ◽  
pp. 125-138 ◽  
Author(s):  
I. V. Goncharenko
Keyword(s):  
Cluster Analysis ◽  
Clustering Algorithm ◽  
Nearest Neighbor ◽  
Clustering Algorithms ◽  
Protein Structures ◽  
Hierarchical Cluster ◽  
Vegetation Classification ◽  
K Nearest Neighbor ◽  
Neighbor Graph ◽  
Nearest Neighbor Graph

In this article we proposed a new method of non-hierarchical cluster analysis using k-nearest-neighbor graph and discussed it with respect to vegetation classification. The method of k-nearest neighbor (k-NN) classification was originally developed in 1951 (Fix, Hodges, 1951). Later a term “k-NN graph” and a few algorithms of k-NN clustering appeared (Cover, Hart, 1967; Brito et al., 1997). In biology k-NN is used in analysis of protein structures and genome sequences. Most of k-NN clustering algorithms build «excessive» graph firstly, so called hypergraph, and then truncate it to subgraphs, just partitioning and coarsening hypergraph. We developed other strategy, the “upward” clustering in forming (assembling consequentially) one cluster after the other. Until today graph-based cluster analysis has not been considered concerning classification of vegetation datasets.

A Novel clustering method based on hybrid K-nearest-neighbor graph

2018 ◽  
Vol 74 ◽  
pp. 1-14 ◽  
Author(s):  
Yikun Qin ◽  
Zhu Liang Yu ◽  
Chang-Dong Wang ◽  
Zhenghui Gu ◽  
Yuanqing Li
Keyword(s):  
Nearest Neighbor ◽  
K Nearest Neighbor ◽  
Clustering Method ◽  
Neighbor Graph ◽  
Nearest Neighbor Graph

Region-Based Graph Learning towards Large Scale Image Annotation

2012 ◽  
pp. 244-260
Author(s):  
Bao Bing-Kun ◽  
Yan Shuicheng
Keyword(s):  
Large Scale ◽  
Nearest Neighbor ◽  
Image Annotation ◽  
Learning Algorithm ◽  
Label Propagation ◽  
Locality Sensitive Hashing ◽  
K Nearest Neighbor ◽  
Neighbor Graph ◽  
Nearest Neighbor Graph ◽  
Modeling Data

Graph-based learning provides a useful approach for modeling data in image annotation problems. In this chapter, the authors introduce how to construct a region-based graph to annotate large scale multi-label images. It has been well recognized that analysis in semantic region level may greatly improve image annotation performance compared to that in whole image level. However, the region level approach increases the data scale to several orders of magnitude and lays down new challenges to most existing algorithms. To this end, each image is firstly encoded as a Bag-of-Regions based on multiple image segmentations. And then, all image regions are constructed into a large k-nearest-neighbor graph with efficient Locality Sensitive Hashing (LSH) method. At last, a sparse and region-aware image-based graph is fed into the multi-label extension of the Entropic graph regularized semi-supervised learning algorithm (Subramanya & Bilmes, 2009). In combination they naturally yield the capability in handling large-scale dataset. Extensive experiments on NUS-WIDE (260k images) and COREL-5k datasets well validate the effectiveness and efficiency of the framework for region-aware and scalable multi-label propagation.

Dimensionality Reduction with Unsupervised Feature Selection and Applying Non-Euclidean Norms for Classification Accuracy

2011 ◽  
pp. 91-109
Author(s):  
Amit Saxena ◽  
John Wang
Keyword(s):  
Classification Accuracy ◽  
Nearest Neighbor ◽  
Fitness Function ◽  
Synthetic Data ◽  
Feature Subset Selection ◽  
Second Phase ◽  
Data Sets ◽  
Feature Subset ◽  
K Nearest Neighbor ◽  
Two Phase

This paper presents a two-phase scheme to select reduced number of features from a dataset using Genetic Algorithm (GA) and testing the classification accuracy (CA) of the dataset with the reduced feature set. In the first phase of the proposed work, an unsupervised approach to select a subset of features is applied. GA is used to select stochastically reduced number of features with Sammon Error as the fitness function. Different subsets of features are obtained. In the second phase, each of the reduced features set is applied to test the CA of the dataset. The CA of a data set is validated using supervised k-nearest neighbor (k-nn) algorithm. The novelty of the proposed scheme is that each reduced feature set obtained in the first phase is investigated for CA using the k-nn classification with different Minkowski metric i.e. non-Euclidean norms instead of conventional Euclidean norm (L2). Final results are presented in the paper with extensive simulations on seven real and one synthetic, data sets. It is revealed from the proposed investigation that taking different norms produces better CA and hence a scope for better feature subset selection.

Sign in / Sign up

Export Citation Format

Share Document