PASCAL Agnostic Learning
vs.
Prior Knowledge
IJCNN07

The challenge is now over. But it remains open for post-challenge submissions!


IMPORTANT: Entries made since February 1st 2007 might be using validation data, now available for training.

serate cubic lssvm

Submitted by reference

LS-SVM with a normalised cubic kernel, model selection via scaled conjugate gradient descent, minimising a smoothed error rate criterion.

Performance Estimates:

ADA - BER = 0.188515 : AUC = 0.899052

GINA - BER = 0.053069 : AUC = 0.987084

HIVA - BER = 0.253982 : AUC = 0.797751

NOVA - BER = 0.055489 : AUC = 0.985058

SYLVA - BER = 0.007697 : AUC = 0.999040

Dataset Balanced Error Area Under Curve  
Train Valid Test Train Valid Test
ada 0.1644 0.2046 0.1889 0.922 0.8726 0.8991 agnostic
gina 0 0.0285 0.047 1 0.9956 0.99 agnostic
hiva 0.0245 0.2832 0.2749 0.9997 0.7512 0.7723 agnostic
nova 0.0004 0.064 0.0538 1 0.9942 0.9888 agnostic
sylva 0.0022 0.0049 0.0081 0.9999 0.9991 0.9988 agnostic
Overall 0.0383 0.1171 0.1145 0.9843 0.9225 0.9298 agnostic