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.

the ugly

Submitted by reference

A combination of reference models, selected according to the leave-one-out BER (this is a bit ugly because the optimal threshold is also detemined by the leave-one-out BER, so it will be rather biased).

The Models:

ADA - press ard lssvm : LOO BER = 0.173187

GINA - press ard lssvm : LOO BER = 0.022994

HIVA - press quadratic lssvm : LOO BER = 0.235784

NOVA - press linear lssvm : LOO BER = 0.042375

SYLVA - press rbf lssvm : LOO BER = 0.005986

Dataset Balanced Error Area Under Curve  
Train Valid Test Train Valid Test
ada 0.1464 0.1806 0.1805 0.9392 0.893 0.9097 agnostic
gina 0 0.0253 0.0333 1 0.9968 0.9939 agnostic
hiva 0.0212 0.2535 0.2733 0.9972 0.7253 0.772 agnostic
nova 0.0004 0.044 0.0481 1 0.9968 0.9894 agnostic
sylva 0.002 0.0049 0.0081 1 0.9991 0.9988 agnostic
Overall 0.034 0.1016 0.1086 0.9873 0.9222 0.9328 agnostic