http://deeplearning.stanford.edu/wiki/index.php?title=%E5%BE%AE%E8%B0%83%E5%A4%9A%E5%B1%82%E8%87%AA%E7%BC%96%E7%A0%81%E7%AE%97%E6%B3%95&feed=atom&action=history
微调多层自编码算法 - Revision history
2024-03-28T22:55:03Z
Revision history for this page on the wiki
MediaWiki 1.16.2
http://deeplearning.stanford.edu/wiki/index.php?title=%E5%BE%AE%E8%B0%83%E5%A4%9A%E5%B1%82%E8%87%AA%E7%BC%96%E7%A0%81%E7%AE%97%E6%B3%95&diff=2344&oldid=prev
Wikiroot at 05:16, 8 April 2013
2013-04-08T05:16:59Z
<p></p>
<table style="background-color: white; color:black;">
<col class='diff-marker' />
<col class='diff-content' />
<col class='diff-marker' />
<col class='diff-content' />
<tr valign='top'>
<td colspan='2' style="background-color: white; color:black;">← Older revision</td>
<td colspan='2' style="background-color: white; color:black;">Revision as of 05:16, 8 April 2013</td>
</tr><tr><td colspan="2" class="diff-lineno">Line 65:</td>
<td colspan="2" class="diff-lineno">Line 65:</td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'>-</td><td style="background: #ffa; color:black; font-size: smaller;"><div>{{<del class="diffchange diffchange-inline">CNN</del>}}</div></td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div>{{<ins class="diffchange diffchange-inline">建立分类用深度网络</ins>}}</div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>{{Languages|Fine-tuning_Stacked_AEs|English}}</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>{{Languages|Fine-tuning_Stacked_AEs|English}}</div></td></tr>
</table>
Wikiroot
http://deeplearning.stanford.edu/wiki/index.php?title=%E5%BE%AE%E8%B0%83%E5%A4%9A%E5%B1%82%E8%87%AA%E7%BC%96%E7%A0%81%E7%AE%97%E6%B3%95&diff=2304&oldid=prev
Kandeng at 04:05, 8 April 2013
2013-04-08T04:05:33Z
<p></p>
<table style="background-color: white; color:black;">
<col class='diff-marker' />
<col class='diff-content' />
<col class='diff-marker' />
<col class='diff-content' />
<tr valign='top'>
<td colspan='2' style="background-color: white; color:black;">← Older revision</td>
<td colspan='2' style="background-color: white; color:black;">Revision as of 04:05, 8 April 2013</td>
</tr><tr><td colspan="2" class="diff-lineno">Line 45:</td>
<td colspan="2" class="diff-lineno">Line 45:</td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>注:我们可以认为输出层softmax分类器是附加上的一层,但是其求导过程需要单独处理。具体地说,网络“最后一层”的特征会进入softmax分类器。所以,第二步中的导数由 <math>\delta^{(n_l)} = - (\nabla_{a^{n_l}}J) \bullet f'(z^{(n_l)})</math> 计算,其中 <math>\nabla J = \theta^T(I-P)</math>。</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>注:我们可以认为输出层softmax分类器是附加上的一层,但是其求导过程需要单独处理。具体地说,网络“最后一层”的特征会进入softmax分类器。所以,第二步中的导数由 <math>\delta^{(n_l)} = - (\nabla_{a^{n_l}}J) \bullet f'(z^{(n_l)})</math> 计算,其中 <math>\nabla J = \theta^T(I-P)</math>。</div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>}}</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>}}</div></td></tr>
<tr><td class='diff-marker'>-</td><td style="background: #ffa; color:black; font-size: smaller;"><div><del style="color: red; font-weight: bold; text-decoration: none;"></del></div></td><td colspan="2"> </td></tr>
<tr><td class='diff-marker'>-</td><td style="background: #ffa; color:black; font-size: smaller;"><div><del style="color: red; font-weight: bold; text-decoration: none;"></del></div></td><td colspan="2"> </td></tr>
<tr><td class='diff-marker'>-</td><td style="background: #ffa; color:black; font-size: smaller;"><div><del style="color: red; font-weight: bold; text-decoration: none;">{{CNN}}</del></div></td><td colspan="2"> </td></tr>
<tr><td class='diff-marker'>-</td><td style="background: #ffa; color:black; font-size: smaller;"><div><del style="color: red; font-weight: bold; text-decoration: none;"></del></div></td><td colspan="2"> </td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td colspan="2" class="diff-lineno">Line 67:</td>
<td colspan="2" class="diff-lineno">Line 63:</td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>崔巍(watsoncui@gmail.com), 余凯(kai.yu.cool@gmail.com), 许利杰(csxulijie@gmail.com)</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>崔巍(watsoncui@gmail.com), 余凯(kai.yu.cool@gmail.com), 许利杰(csxulijie@gmail.com)</div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins style="color: red; font-weight: bold; text-decoration: none;"></ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins style="color: red; font-weight: bold; text-decoration: none;"></ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins style="color: red; font-weight: bold; text-decoration: none;">{{CNN}}</ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins style="color: red; font-weight: bold; text-decoration: none;"></ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins style="color: red; font-weight: bold; text-decoration: none;"></ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins style="color: red; font-weight: bold; text-decoration: none;">{{Languages|Fine-tuning_Stacked_AEs|English}}</ins></div></td></tr>
</table>
Kandeng
http://deeplearning.stanford.edu/wiki/index.php?title=%E5%BE%AE%E8%B0%83%E5%A4%9A%E5%B1%82%E8%87%AA%E7%BC%96%E7%A0%81%E7%AE%97%E6%B3%95&diff=2231&oldid=prev
Kandeng at 15:41, 5 April 2013
2013-04-05T15:41:52Z
<p></p>
<table style="background-color: white; color:black;">
<col class='diff-marker' />
<col class='diff-content' />
<col class='diff-marker' />
<col class='diff-content' />
<tr valign='top'>
<td colspan='2' style="background-color: white; color:black;">← Older revision</td>
<td colspan='2' style="background-color: white; color:black;">Revision as of 15:41, 5 April 2013</td>
</tr><tr><td colspan="2" class="diff-lineno">Line 45:</td>
<td colspan="2" class="diff-lineno">Line 45:</td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>注:我们可以认为输出层softmax分类器是附加上的一层,但是其求导过程需要单独处理。具体地说,网络“最后一层”的特征会进入softmax分类器。所以,第二步中的导数由 <math>\delta^{(n_l)} = - (\nabla_{a^{n_l}}J) \bullet f'(z^{(n_l)})</math> 计算,其中 <math>\nabla J = \theta^T(I-P)</math>。</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>注:我们可以认为输出层softmax分类器是附加上的一层,但是其求导过程需要单独处理。具体地说,网络“最后一层”的特征会进入softmax分类器。所以,第二步中的导数由 <math>\delta^{(n_l)} = - (\nabla_{a^{n_l}}J) \bullet f'(z^{(n_l)})</math> 计算,其中 <math>\nabla J = \theta^T(I-P)</math>。</div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>}}</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>}}</div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins style="color: red; font-weight: bold; text-decoration: none;"></ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins style="color: red; font-weight: bold; text-decoration: none;"></ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins style="color: red; font-weight: bold; text-decoration: none;">{{CNN}}</ins></div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>==中英文对照==</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>==中英文对照==</div></td></tr>
<tr><td class='diff-marker'>-</td><td style="background: #ffa; color:black; font-size: smaller;"><div><del style="color: red; font-weight: bold; text-decoration: none;">stacked autoencoder 栈式自编码神经网络(可以考虑翻译为“多层自动编码机”或“多层自动编码神经网络”)</del></div></td><td colspan="2"> </td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'>-</td><td style="background: #ffa; color:black; font-size: smaller;"><div><del class="diffchange diffchange-inline">fine tuning 微调</del></div></td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins class="diffchange diffchange-inline">:栈式自编码神经网络(可以考虑翻译为“多层自动编码机”或“多层自动编码神经网络”) Stacked autoencoder </ins></div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'>-</td><td style="background: #ffa; color:black; font-size: smaller;"><div><del class="diffchange diffchange-inline">Backpropagation Algorithm 反向传播算法</del></div></td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins class="diffchange diffchange-inline">:微调 Fine tuning </ins></div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'>-</td><td style="background: #ffa; color:black; font-size: smaller;"><div><del class="diffchange diffchange-inline">feedforward pass 前馈传递</del></div></td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins class="diffchange diffchange-inline">:反向传播算法 Backpropagation Algorithm </ins></div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'>-</td><td style="background: #ffa; color:black; font-size: smaller;"><div><del class="diffchange diffchange-inline">activation 激活值 (可以考虑翻译为“激励响应”或“响应”)</del></div></td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins class="diffchange diffchange-inline">:前馈传递 feedforward pass </ins></div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins style="color: red; font-weight: bold; text-decoration: none;">:激活值 (可以考虑翻译为“激励响应”或“响应”) activation </ins></div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>==中文译者==</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>==中文译者==</div></td></tr>
<tr><td class='diff-marker'>-</td><td style="background: #ffa; color:black; font-size: smaller;"><div><del class="diffchange diffchange-inline">初译:</del>@<del class="diffchange diffchange-inline">太二真人 一审:</del>@<del class="diffchange diffchange-inline">余凯_西二旗民工 终审:</del>@<del class="diffchange diffchange-inline">JerryLead</del></div></td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div> </div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins class="diffchange diffchange-inline">崔巍(watsoncui</ins>@<ins class="diffchange diffchange-inline">gmail.com), 余凯(kai.yu.cool</ins>@<ins class="diffchange diffchange-inline">gmail.com), 许利杰(csxulijie</ins>@<ins class="diffchange diffchange-inline">gmail.com)</ins></div></td></tr>
</table>
Kandeng
http://deeplearning.stanford.edu/wiki/index.php?title=%E5%BE%AE%E8%B0%83%E5%A4%9A%E5%B1%82%E8%87%AA%E7%BC%96%E7%A0%81%E7%AE%97%E6%B3%95&diff=2148&oldid=prev
Kandeng at 13:11, 30 March 2013
2013-03-30T13:11:57Z
<p></p>
<table style="background-color: white; color:black;">
<col class='diff-marker' />
<col class='diff-content' />
<col class='diff-marker' />
<col class='diff-content' />
<tr valign='top'>
<td colspan='2' style="background-color: white; color:black;">← Older revision</td>
<td colspan='2' style="background-color: white; color:black;">Revision as of 13:11, 30 March 2013</td>
</tr><tr><td colspan="2" class="diff-lineno">Line 45:</td>
<td colspan="2" class="diff-lineno">Line 45:</td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>注:我们可以认为输出层softmax分类器是附加上的一层,但是其求导过程需要单独处理。具体地说,网络“最后一层”的特征会进入softmax分类器。所以,第二步中的导数由 <math>\delta^{(n_l)} = - (\nabla_{a^{n_l}}J) \bullet f'(z^{(n_l)})</math> 计算,其中 <math>\nabla J = \theta^T(I-P)</math>。</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>注:我们可以认为输出层softmax分类器是附加上的一层,但是其求导过程需要单独处理。具体地说,网络“最后一层”的特征会进入softmax分类器。所以,第二步中的导数由 <math>\delta^{(n_l)} = - (\nabla_{a^{n_l}}J) \bullet f'(z^{(n_l)})</math> 计算,其中 <math>\nabla J = \theta^T(I-P)</math>。</div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>}}</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>}}</div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins style="color: red; font-weight: bold; text-decoration: none;"></ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins style="color: red; font-weight: bold; text-decoration: none;"></ins></div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>==中英文对照==</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>==中英文对照==</div></td></tr>
</table>
Kandeng
http://deeplearning.stanford.edu/wiki/index.php?title=%E5%BE%AE%E8%B0%83%E5%A4%9A%E5%B1%82%E8%87%AA%E7%BC%96%E7%A0%81%E7%AE%97%E6%B3%95&diff=2147&oldid=prev
Kandeng: /* 使用反向传播法进行微调 */
2013-03-30T13:11:36Z
<p><span class="autocomment">使用反向传播法进行微调</span></p>
<table style="background-color: white; color:black;">
<col class='diff-marker' />
<col class='diff-content' />
<col class='diff-marker' />
<col class='diff-content' />
<tr valign='top'>
<td colspan='2' style="background-color: white; color:black;">← Older revision</td>
<td colspan='2' style="background-color: white; color:black;">Revision as of 13:11, 30 March 2013</td>
</tr><tr><td colspan="2" class="diff-lineno">Line 43:</td>
<td colspan="2" class="diff-lineno">Line 43:</td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>{{Quote|</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>{{Quote|</div></td></tr>
<tr><td class='diff-marker'>-</td><td style="background: #ffa; color:black; font-size: smaller;"><div><del class="diffchange diffchange-inline">注:我们可以认为输出层softmax分类器是附加上的一层,但是其求导过程需要分别处理。具体地说,网络“最后一层”的特征会进入softmax分类器。所以,第二步中的导数由 </del><math>\delta^{(n_l)} = - (\nabla_{a^{n_l}}J) \bullet f'(z^{(n_l)})</math> 计算,其中 <math>\nabla J = \theta^T(I-P)</math>。</div></td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins class="diffchange diffchange-inline">注:我们可以认为输出层softmax分类器是附加上的一层,但是其求导过程需要单独处理。具体地说,网络“最后一层”的特征会进入softmax分类器。所以,第二步中的导数由 </ins><math>\delta^{(n_l)} = - (\nabla_{a^{n_l}}J) \bullet f'(z^{(n_l)})</math> 计算,其中 <math>\nabla J = \theta^T(I-P)</math>。</div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>}}</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>}}</div></td></tr>
<tr><td class='diff-marker'>-</td><td style="background: #ffa; color:black; font-size: smaller;"><div><del style="color: red; font-weight: bold; text-decoration: none;"></del></div></td><td colspan="2"> </td></tr>
<tr><td class='diff-marker'>-</td><td style="background: #ffa; color:black; font-size: smaller;"><div><del style="color: red; font-weight: bold; text-decoration: none;"></del></div></td><td colspan="2"> </td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>==中英文对照==</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>==中英文对照==</div></td></tr>
</table>
Kandeng
http://deeplearning.stanford.edu/wiki/index.php?title=%E5%BE%AE%E8%B0%83%E5%A4%9A%E5%B1%82%E8%87%AA%E7%BC%96%E7%A0%81%E7%AE%97%E6%B3%95&diff=2146&oldid=prev
Kandeng at 13:11, 30 March 2013
2013-03-30T13:11:11Z
<p></p>
<table style="background-color: white; color:black;">
<col class='diff-marker' />
<col class='diff-content' />
<col class='diff-marker' />
<col class='diff-content' />
<tr valign='top'>
<td colspan='2' style="background-color: white; color:black;">← Older revision</td>
<td colspan='2' style="background-color: white; color:black;">Revision as of 13:11, 30 March 2013</td>
</tr><tr><td colspan="2" class="diff-lineno">Line 45:</td>
<td colspan="2" class="diff-lineno">Line 45:</td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>注:我们可以认为输出层softmax分类器是附加上的一层,但是其求导过程需要分别处理。具体地说,网络“最后一层”的特征会进入softmax分类器。所以,第二步中的导数由 <math>\delta^{(n_l)} = - (\nabla_{a^{n_l}}J) \bullet f'(z^{(n_l)})</math> 计算,其中 <math>\nabla J = \theta^T(I-P)</math>。</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>注:我们可以认为输出层softmax分类器是附加上的一层,但是其求导过程需要分别处理。具体地说,网络“最后一层”的特征会进入softmax分类器。所以,第二步中的导数由 <math>\delta^{(n_l)} = - (\nabla_{a^{n_l}}J) \bullet f'(z^{(n_l)})</math> 计算,其中 <math>\nabla J = \theta^T(I-P)</math>。</div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>}}</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>}}</div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins style="color: red; font-weight: bold; text-decoration: none;"></ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins style="color: red; font-weight: bold; text-decoration: none;"></ins></div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>==中英文对照==</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>==中英文对照==</div></td></tr>
<tr><td colspan="2" class="diff-lineno">Line 56:</td>
<td colspan="2" class="diff-lineno">Line 58:</td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>activation 激活值 (可以考虑翻译为“激励响应”或“响应”)</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>activation 激活值 (可以考虑翻译为“激励响应”或“响应”)</div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins style="color: red; font-weight: bold; text-decoration: none;"></ins></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins style="color: red; font-weight: bold; text-decoration: none;"></ins></div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>==中文译者==</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>==中文译者==</div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>初译:@太二真人 一审:@余凯_西二旗民工 终审:@JerryLead</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>初译:@太二真人 一审:@余凯_西二旗民工 终审:@JerryLead</div></td></tr>
</table>
Kandeng
http://deeplearning.stanford.edu/wiki/index.php?title=%E5%BE%AE%E8%B0%83%E5%A4%9A%E5%B1%82%E8%87%AA%E7%BC%96%E7%A0%81%E7%AE%97%E6%B3%95&diff=2145&oldid=prev
Kandeng at 13:10, 30 March 2013
2013-03-30T13:10:55Z
<p></p>
<a href="http://deeplearning.stanford.edu/wiki/index.php?title=%E5%BE%AE%E8%B0%83%E5%A4%9A%E5%B1%82%E8%87%AA%E7%BC%96%E7%A0%81%E7%AE%97%E6%B3%95&diff=2145&oldid=2144">Show changes</a>
Kandeng
http://deeplearning.stanford.edu/wiki/index.php?title=%E5%BE%AE%E8%B0%83%E5%A4%9A%E5%B1%82%E8%87%AA%E7%BC%96%E7%A0%81%E7%AE%97%E6%B3%95&diff=2144&oldid=prev
Kandeng: /* 使用反向传播法进行微调 */
2013-03-30T13:09:39Z
<p><span class="autocomment">使用反向传播法进行微调</span></p>
<table style="background-color: white; color:black;">
<col class='diff-marker' />
<col class='diff-content' />
<col class='diff-marker' />
<col class='diff-content' />
<tr valign='top'>
<td colspan='2' style="background-color: white; color:black;">← Older revision</td>
<td colspan='2' style="background-color: white; color:black;">Revision as of 13:09, 30 March 2013</td>
</tr><tr><td colspan="2" class="diff-lineno">Line 43:</td>
<td colspan="2" class="diff-lineno">Line 43:</td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>{{Quote|</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>{{Quote|</div></td></tr>
<tr><td class='diff-marker'>-</td><td style="background: #ffa; color:black; font-size: smaller;"><div><del class="diffchange diffchange-inline">注:我们可以认为输出层softmax分类器是附加上的一层,但是其推导需要分别处理。具体地说,网络“最后一层”的特征会进入softmax分类器。所以,第二步中的导数由 </del><math>\delta^{(n_l)} = - (\nabla_{a^{n_l}}J) \bullet f'(z^{(n_l)})</math> 计算,其中 <math>\nabla J = \theta^T(I-P)</math>。</div></td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins class="diffchange diffchange-inline">注:我们可以认为输出层softmax分类器是附加上的一层,但是其求导过程需要分别处理。具体地说,网络“最后一层”的特征会进入softmax分类器。所以,第二步中的导数由 </ins><math>\delta^{(n_l)} = - (\nabla_{a^{n_l}}J) \bullet f'(z^{(n_l)})</math> 计算,其中 <math>\nabla J = \theta^T(I-P)</math>。</div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>}}</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>}}</div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
</table>
Kandeng
http://deeplearning.stanford.edu/wiki/index.php?title=%E5%BE%AE%E8%B0%83%E5%A4%9A%E5%B1%82%E8%87%AA%E7%BC%96%E7%A0%81%E7%AE%97%E6%B3%95&diff=2143&oldid=prev
Kandeng: /* 使用反向传播法进行微调 */
2013-03-30T13:07:51Z
<p><span class="autocomment">使用反向传播法进行微调</span></p>
<table style="background-color: white; color:black;">
<col class='diff-marker' />
<col class='diff-content' />
<col class='diff-marker' />
<col class='diff-content' />
<tr valign='top'>
<td colspan='2' style="background-color: white; color:black;">← Older revision</td>
<td colspan='2' style="background-color: white; color:black;">Revision as of 13:07, 30 March 2013</td>
</tr><tr><td colspan="2" class="diff-lineno">Line 19:</td>
<td colspan="2" class="diff-lineno">Line 19:</td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>= - (\nabla_{a^{n_l}}J) \bullet f'(z^{(n_l)})</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>= - (\nabla_{a^{n_l}}J) \bullet f'(z^{(n_l)})</div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>\end{align}</math></div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>\end{align}</math></div></td></tr>
<tr><td colspan="2"> </td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div><ins style="color: red; font-weight: bold; text-decoration: none;"></ins></div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>::(当使用softmax分类器时,softmax层满足:<math>\nabla J = \theta^T(I-P)</math>,其中 <math>\textstyle I</math> 为输入数据对应的类别标签,<math>\textstyle P</math> 为条件概率向量。)</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>::(当使用softmax分类器时,softmax层满足:<math>\nabla J = \theta^T(I-P)</math>,其中 <math>\textstyle I</math> 为输入数据对应的类别标签,<math>\textstyle P</math> 为条件概率向量。)</div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
</table>
Kandeng
http://deeplearning.stanford.edu/wiki/index.php?title=%E5%BE%AE%E8%B0%83%E5%A4%9A%E5%B1%82%E8%87%AA%E7%BC%96%E7%A0%81%E7%AE%97%E6%B3%95&diff=2142&oldid=prev
Kandeng: /* 使用反向传播法微调 */
2013-03-30T13:07:04Z
<p><span class="autocomment">使用反向传播法微调</span></p>
<table style="background-color: white; color:black;">
<col class='diff-marker' />
<col class='diff-content' />
<col class='diff-marker' />
<col class='diff-content' />
<tr valign='top'>
<td colspan='2' style="background-color: white; color:black;">← Older revision</td>
<td colspan='2' style="background-color: white; color:black;">Revision as of 13:07, 30 March 2013</td>
</tr><tr><td colspan="2" class="diff-lineno">Line 6:</td>
<td colspan="2" class="diff-lineno">Line 6:</td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>幸运的是,实施微调栈式自编码神经网络所需的工具都已齐备!为了在每次迭代中计算所有层的梯度,我们需要使用稀疏自动编码一节中讨论的[[反向传播算法]]。因为反向传播算法可以延伸应用到任意多层,所以事实上,该算法对任意多层的栈式自编码神经网络都适用。</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>幸运的是,实施微调栈式自编码神经网络所需的工具都已齐备!为了在每次迭代中计算所有层的梯度,我们需要使用稀疏自动编码一节中讨论的[[反向传播算法]]。因为反向传播算法可以延伸应用到任意多层,所以事实上,该算法对任意多层的栈式自编码神经网络都适用。</div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'>-</td><td style="background: #ffa; color:black; font-size: smaller;"><div>==<del class="diffchange diffchange-inline">使用反向传播法微调</del>==</div></td><td class='diff-marker'>+</td><td style="background: #cfc; color:black; font-size: smaller;"><div>==<ins class="diffchange diffchange-inline">使用反向传播法进行微调</ins>==</div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>为方便读者,以下我们简要描述如何实施反向传播算法:</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>为方便读者,以下我们简要描述如何实施反向传播算法:</div></td></tr>
<tr><td colspan="2" class="diff-lineno">Line 44:</td>
<td colspan="2" class="diff-lineno">Line 44:</td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>注:我们可以认为输出层softmax分类器是附加上的一层,但是其推导需要分别处理。具体地说,网络“最后一层”的特征会进入softmax分类器。所以,第二步中的导数由 <math>\delta^{(n_l)} = - (\nabla_{a^{n_l}}J) \bullet f'(z^{(n_l)})</math> 计算,其中 <math>\nabla J = \theta^T(I-P)</math>。</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>注:我们可以认为输出层softmax分类器是附加上的一层,但是其推导需要分别处理。具体地说,网络“最后一层”的特征会进入softmax分类器。所以,第二步中的导数由 <math>\delta^{(n_l)} = - (\nabla_{a^{n_l}}J) \bullet f'(z^{(n_l)})</math> 计算,其中 <math>\nabla J = \theta^T(I-P)</math>。</div></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>}}</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>}}</div></td></tr>
<tr><td class='diff-marker'>-</td><td style="background: #ffa; color:black; font-size: smaller;"><div><del style="color: red; font-weight: bold; text-decoration: none;"></del></div></td><td colspan="2"> </td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"></td></tr>
<tr><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>==中英文对照==</div></td><td class='diff-marker'> </td><td style="background: #eee; color:black; font-size: smaller;"><div>==中英文对照==</div></td></tr>
</table>
Kandeng