DocumentCode :
2407
Title :
Gentle Nearest Neighbors Boosting over Proper Scoring Rules
Author :
Nock, Richard ; Ali, Wafa Bel Haj ; D´Ambrosio, Roberto ; Nielsen, Frank ; Barlaud, Michel
Author_Institution :
Dept. of Sci. Interfacultaire, Ceregmia-Univ. Antilles-Guyane, Guyane, France
Volume :
37
Issue :
1
fYear :
2015
fDate :
Jan. 2015
Firstpage :
80
Lastpage :
93
Abstract :
Tailoring nearest neighbors algorithms to boosting is an important problem. Recent papers study an approach, UNN, which provably minimizes particular convex surrogates under weak assumptions. However, numerical issues make it necessary to experimentally tweak parts of the UNN algorithm, at the possible expense of the algorithm´s convergence and performance. In this paper, we propose a lightweight Newton-Raphson alternative optimizing proper scoring rules from a very broad set, and establish formal convergence rates under the boosting framework that compete with those known for UNN. To the best of our knowledge, no such boosting-compliant convergence rates were previously known in the popular Gentle Adaboost´s lineage. We provide experiments on a dozen domains, including Caltech and SUN computer vision databases, comparing our approach to major families including support vector machines, (Ada)boosting and stochastic gradient descent. They support three major conclusions: (i) GNNB significantly outperforms UNN, in terms of convergence rate and quality of the outputs, (ii) GNNB performs on par with or better than computationally intensive large margin approaches, (iii) on large domains that rule out those latter approaches for computational reasons, GNNB provides a simple and competitive contender to stochastic gradient descent. Experiments include a divide-and-conquer improvement of GNNB exploiting the link with proper scoring rules optimization.
Keywords :
Newton-Raphson method; convergence; learning (artificial intelligence); pattern classification; Adaboosting; Caltech; GNNB; Newton-Raphson alternative optimizing proper scoring rules; SUN computer vision databases; UNN; boosting-compliant convergence rates; classifiers; formal convergence rates; gentle nearest neighbors boosting; stochastic gradient descent; support vector machines; universal nearest neighbors; Boosting; Convergence; Estimation; Logistics; Minimization; Optimization; Vectors; Boosting; nearest neighbors; proper scoring rules;
fLanguage :
English
Journal_Title :
Pattern Analysis and Machine Intelligence, IEEE Transactions on
Publisher :
ieee
ISSN :
0162-8828
Type :
jour
DOI :
10.1109/TPAMI.2014.2307877
Filename :
6747340
Link To Document :
بازگشت