Distance Metric Learning for Large Margin Nearest Neighbor Classification

Distance Metric Learning for Large Margin    Nearest Neighbor Classification

key words:

Nahanalobis distance:  not dependent on the scale of measurement. 

knn classification can be significantly improved by using a distance metric learned from labeled examples.

many researches -> knn classification can be significantly improved by using a distance metric learned from labeled examples. 

approach similar to svm

large margin nearest neighbor LMNN classification

linear transformation that optimizes knn

cost funtion

penalizes large distances between each input and its target neighbors and penalizes small distances between each input and all other inputs that do not share the same label. 

matlab implementation online. 

related work

semidefinite programming

neighborhood component analisis. 

relevant component analisis.

PCA

kernelizing the algorithm will  improve it