Mercurial > ift6266

comparison writeup/nips2010_submission.tex @ 536:5157a5830125

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
One comma
author Dumitru Erhan <dumitru.erhan@gmail.com>
date Tue, 01 Jun 2010 18:28:09 -0700
parents 85f2337d47d2
children 47894d0ecbde
comparison
equal deleted inserted replaced
531:85f2337d47d2 536:5157a5830125
689 In the original self-taught learning framework~\citep{RainaR2007}, the 689 In the original self-taught learning framework~\citep{RainaR2007}, the
690 out-of-sample examples were used as a source of unsupervised data, and 690 out-of-sample examples were used as a source of unsupervised data, and
691 experiments showed its positive effects in a \emph{limited labeled data} 691 experiments showed its positive effects in a \emph{limited labeled data}
692 scenario. However, many of the results by \citet{RainaR2007} (who used a 692 scenario. However, many of the results by \citet{RainaR2007} (who used a
693 shallow, sparse coding approach) suggest that the relative gain of self-taught 693 shallow, sparse coding approach) suggest that the relative gain of self-taught
694 learning diminishes as the number of labeled examples increases, (essentially, 694 learning diminishes as the number of labeled examples increases (essentially,
695 a ``diminishing returns'' scenario occurs). We note instead that, for deep 695 a ``diminishing returns'' scenario occurs). We note instead that, for deep
696 architectures, our experiments show that such a positive effect is accomplished 696 architectures, our experiments show that such a positive effect is accomplished
697 even in a scenario with a \emph{very large number of labeled examples}. 697 even in a scenario with a \emph{very large number of labeled examples}.
698 698
699 Why would deep learners benefit more from the self-taught learning framework? 699 Why would deep learners benefit more from the self-taught learning framework?