ν-MCD approach to novelty detection

Published on Feb 25, 20073201 Views

Workshop on ML, SVM and Large Scale Optimization 2005 - Thurnau

Novelty Detection Based on Information Matrix00:02

Motivation00:38

Novelty Detection: definitions01:34

Example 103:10

Training set: Researches in Machine Learning03:44

Possible approaches04:43

Experimental design06:22

Equivalence Theorem08:11

Minimum Covering Ellipsoid09:23

Minimum Covering Ellipsoid in 10:58

Optimization12:16

Titterington’s algorithm: Experiment (1)14:01

Proposed approach 14:55

Lagrangian17:28

Dual problem, experimental design17:52

What is outlier?20:09

Rousseeuw's MCD method20:37

Experiment (2)22:39

Experiment (3)24:07

Kernel Principal Component Analysis (Scholkopf, Smola, Muller, 1996) 24:43

Properties of non-robust MCE using KPCA25:51

Experiment (4)26:49

Experiment (5)27:44

KPCA+Rousseeuw's MCD method28:50

The minimum covering sphere problem and S-optimum experimental design 29:55

Simple algebra31:02

Illustrations of novelty detection methods31:15

Relations to Tax and Scholkopf methods 32:29

Potential Applications34:41

Simulation and Testing35:06

Conclusions35:19

Future work36:22

Acknowledgement37:07

Thank you! 37:36