A Boosting Approach to Supervised/Semi-Supervised Learning a Mahalanobis Distance Metric for the Nearest-Neighbor Classification

View Statistics Email Alert RSS Feed

Information

Details

Project title

Code/計畫編號

NSC99-2221-E019-036

Translated Name/計畫中文名

基於最短距離分類法與提昇法之監督式/半監督式學習Mahalanobis距離度量演算法

Project Coordinator/計畫主持人

Chin-Chun Chang

Funding Organization/主管機關

National Science and Technology Council

Department/Unit

Department of Computer Science and Engineering

Website

https://www.grb.gov.tw/search/planDetail?id=2116297

Year

2010

Bugetid/研究經費

435千元

ResearchField/研究領域

資訊科學--軟體

"在輸入空間決定一個恰當的距離度量，是一個圖形識別與機器學習的基本問題。許多電腦視覺的應用都需要一個好的距離度量。在這個兩年計畫裡，我們將發展學習Mahalanobis距離度量的監督式/半監督式演算法。尤其，我們將採用提昇法(the boosting algorithm)及使用一個針對最短距離分類法設計的新穎失敗函數來發展這個演算法。本兩年計畫的主題分別如下：  第一年我們將針對最短距離分類法來設計一個新穎的失敗函數。利用此失敗函數，我們將使用提昇法來發展一個全監督式Mahalanobis距離度量學習演算法。  第二年我們將擴展第一年所發展的演算法至半監督式問題。我們將使用圖形式標籤傳播法(the graph-based label propagation algorithm)來從已標示樣本傳播標籤資訊至未標示樣本，並結合至第一年所發展的演算法裡。由於我們將使用提昇法與針對最短距離分類法所設計的失敗函數，此演算法所學習到的距離度量對最短距離分類法的效果可期。""Determining a proper distance metric in the input space is one of the fundamental problems in pattern recognition and machine learning. Many applications of computer vision, need good distance metrics. In this two-year project, we shall focus on developing algorithms for learning a Mahalanobis distance metric from training samples. In particular, we shall adopt the boosting technique with a novel loss function for the nearest-neighbor classification to develop such algorithms for supervised problems, and semi-supervised problems. The topics of this two-year project are as follows.  In the first year, we shall develop a boosting algorithm for supervised learning a Mahalanobis distance metric based on a novel nearest-neighbor classification loss.  In the second year, we shall extend the algorithm developed in the first year to the semi-supervised problem. We shall adopt a graph-based label propagation algorithm to propagate label information from labeled samples to unlabeled samples. In addition, we shall integrate this label propagation algorithm into the algorithm developed in the first year. Since the algorithms to be developed are based on the boosting technique and a loss function for the nearest-neighbor classification, positive effect of the learned distance-metric on the nearest-neighbor classification could be expected."

Keyword(s)

距離學習
假說式邊沿
提升法
Distance Metric Learning
Hypothesis margins
Boosting Approaches

DSpace CRIS

A Boosting Approach to Supervised/Semi-Supervised Learning a Mahalanobis Distance Metric for the Nearest-Neighbor Classification

Details

Description