Mercurial > ift6266

comparison writeup/techreport.tex @ 461:9609c5cf9b6b

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
lit. results
author Yoshua Bengio <bengioy@iro.umontreal.ca>
date Fri, 28 May 2010 07:48:14 -0600
parents fe292653a0f8
children f59af1648d83
comparison
equal deleted inserted replaced
460:fe292653a0f8 461:9609c5cf9b6b
378 (MLP=Multi-Layer Perceptron). The models shown are all trained using perturbed data (NISTP or P07) 378 (MLP=Multi-Layer Perceptron). The models shown are all trained using perturbed data (NISTP or P07)
379 and using a validation set to select hyper-parameters and other training choices. 379 and using a validation set to select hyper-parameters and other training choices.
380 \{SDA,MLP\}0 are trained on NIST, 380 \{SDA,MLP\}0 are trained on NIST,
381 \{SDA,MLP\}1 are trained on NISTP, and \{SDA,MLP\}2 are trained on P07. 381 \{SDA,MLP\}1 are trained on NISTP, and \{SDA,MLP\}2 are trained on P07.
382 The human error rate on digits is a lower bound because it does not count digits that were 382 The human error rate on digits is a lower bound because it does not count digits that were
383 recognized as letters.} 383 recognized as letters. For comparison, the results found in the literature
384 on NIST digits classification using the same test set are included.}
384 \label{tab:sda-vs-mlp-vs-humans} 385 \label{tab:sda-vs-mlp-vs-humans}
385 \begin{center} 386 \begin{center}
386 \begin{tabular}{|l|r|r|r|r|} \hline 387 \begin{tabular}{|l|r|r|r|r|} \hline
387 & NIST test & NISTP test & P07 test & NIST test digits \\ \hline 388 & NIST test & NISTP test & P07 test & NIST test digits \\ \hline
388 Humans& 18.2\% $\pm$.1\% & 39.4\%$\pm$.1\% & 46.9\%$\pm$.1\% & $>1.1\%$ \\ \hline 389 Humans& 18.2\% $\pm$.1\% & 39.4\%$\pm$.1\% & 46.9\%$\pm$.1\% & $>1.1\%$ \\ \hline
390 SDA1 & 17.1\% $\pm$.13\% & 29.7\%$\pm$.3\% & 29.7\%$\pm$.3\% & 1.4\% $\pm$.1\%\\ \hline 391 SDA1 & 17.1\% $\pm$.13\% & 29.7\%$\pm$.3\% & 29.7\%$\pm$.3\% & 1.4\% $\pm$.1\%\\ \hline
391 SDA2 & 18.7\% $\pm$.13\% & 33.6\%$\pm$.3\% & 39.9\%$\pm$.17\% & 1.7\% $\pm$.1\%\\ \hline 392 SDA2 & 18.7\% $\pm$.13\% & 33.6\%$\pm$.3\% & 39.9\%$\pm$.17\% & 1.7\% $\pm$.1\%\\ \hline
392 MLP0 & 24.2\% $\pm$.15\% & 68.8\%$\pm$.33\% & 78.70\%$\pm$.14\% & 3.45\% $\pm$.15\% \\ \hline 393 MLP0 & 24.2\% $\pm$.15\% & 68.8\%$\pm$.33\% & 78.70\%$\pm$.14\% & 3.45\% $\pm$.15\% \\ \hline
393 MLP1 & 23.0\% $\pm$.15\% & 41.8\%$\pm$.35\% & 90.4\%$\pm$.1\% & 3.85\% $\pm$.16\% \\ \hline 394 MLP1 & 23.0\% $\pm$.15\% & 41.8\%$\pm$.35\% & 90.4\%$\pm$.1\% & 3.85\% $\pm$.16\% \\ \hline
394 MLP2 & 24.3\% $\pm$.15\% & 46.0\%$\pm$.35\% & 54.7\%$\pm$.17\% & 4.85\% $\pm$.18\% \\ \hline 395 MLP2 & 24.3\% $\pm$.15\% & 46.0\%$\pm$.35\% & 54.7\%$\pm$.17\% & 4.85\% $\pm$.18\% \\ \hline
396 [5] & & & & 4.95\% $\pm$.18\% \\ \hline
397 [2] & & & & 3.71\% $\pm$.16\% \\ \hline
398 [3] & & & & 2.4\% $\pm$.13\% \\ \hline
399 [4] & & & & 2.1\% $\pm$.12\% \\ \hline
395 \end{tabular} 400 \end{tabular}
396 \end{center} 401 \end{center}
397 \end{table} 402 \end{table}
398 403
399 \subsection{Perturbed Training Data More Helpful for SDAE} 404 \subsection{Perturbed Training Data More Helpful for SDAE}