Authors
Maya R Gupta, Samy Bengio, Jason Weston
Publication date
2014/4/1
Journal
Journal of Machine Learning Research
Volume
15
Issue
1
Pages
1461-1492
Description
Abstract Classification problems with thousands or more classes often have a large range of
classconfusabilities, and we show that the more-confusable classes add more noise to the
empirical loss that is minimized during training. We propose an online solution that reduces
the effect of highly confusable classes in training the classifier parameters, and focuses the
training on pairs of classes that are easier to differentiate at any given time in the training.
We also show that the adagrad method, recently proposed for automatically decreasing ...
classconfusabilities, and we show that the more-confusable classes add more noise to the
empirical loss that is minimized during training. We propose an online solution that reduces
the effect of highly confusable classes in training the classifier parameters, and focuses the
training on pairs of classes that are easier to differentiate at any given time in the training.
We also show that the adagrad method, recently proposed for automatically decreasing ...
Total citations
Scholar articles
MR Gupta, S Bengio, J Weston - Journal of Machine Learning Research, 2014
Dates and citation counts are estimated and are determined automatically by a computer program.