Baidu
map

递归神经网络(RNNs)在机器深度学习中的奇妙应用

2015-09-04 MedSci MedSci原创

递归神经网络(RNNs)有一些不可思议的地方。从花几十分钟训练我的第一个婴儿模型(相当随意挑选的超参数)开始,到训练出能够针对图像给出有意义描述的模型。有些时候,模型对于输出结果质量的简单程度的比例,会与你的期望相差甚远,而这还仅仅是其中一点。有如此令人震惊结果,许多人认为是因为RNNs非常难训练(事实上,通过多次试验,我得出了相反的结论)。一年前:我一直在训练RNNs,我多次见证了它们的强大

递归神经网络(RNNs)有一些不可思议的地方。从花几十分钟训练我的第一个婴儿模型(相当随意挑选的超参数)开始,到训练出能够针对图像给出有意义描述的模型。有些时候,模型对于输出结果质量的简单程度的比例,会与你的期望相差甚远,而这还仅仅是其中一点。有如此令人震惊结果,许多人认为是因为RNNs非常难训练(事实上,通过多次试验,我得出了相反的结论)。一年前:我一直在训练RNNs,我多次见证了它们的强大的功能和鲁棒性,而且它们的输出结果同样让我感到有趣。这篇文章将会给你展现它不可思议的地方。 顺便说句,在讲述这篇文章的同时,同样会将代码上传到 Github 上,这样你就可以基于多层LSTMs来训练字符级语言模型。你向它输入大量的文本,它会学习并产生类似的文本。你也可以用它来重新运行我下面的代码。但是我们正在不断超越自己;那么RNNs究竟是什么呢? 递归神经网络 序列。你可能会问:是什么让递归神经网络如此特殊?Vanilla神经网络(卷积网络也一样)最大的局限之处就是它们API的局限性:它们将固定大小的向量作为输入(比如一张图片),然后输出一个固定大小的向量(比如不同分类的概率

版权声明:
本网站所有内容来源注明为“梅斯医学”或“MedSci原创”的文字、图片和音视频资料,版权均属于梅斯医学所有。非经授权,任何媒体、网站或个人不得转载,授权转载时须注明来源为“梅斯医学”。其它来源的文章系转载文章,或“梅斯号”自媒体发布的文章,仅系出于传递更多信息之目的,本站仅负责审核内容合规,其内容不代表本站立场,本站不负责内容的准确性和版权。如果存在侵权、或不希望被转载的媒体或个人可与我们联系,我们将立即进行删除处理。
在此留言
评论区 (3)
#插入话题
  1. [GetPortalCommentsPageByObjectIdResponse(id=1582921, encodeId=2792158292163, content=<a href='/topic/show?id=f6e1611840d' target=_blank style='color:#2F92EE;'>#机器#</a>, beContent=null, objectType=article, channel=null, level=null, likeNumber=0, replyNumber=0, topicName=null, topicId=null, topicList=[TopicDto(id=61184, encryptionId=f6e1611840d, topicName=机器)], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=13dc16973127, createdName=ms6279672939590805, createdTime=Sun Sep 06 01:26:00 CST 2015, time=2015-09-06, status=1, ipAttribution=), GetPortalCommentsPageByObjectIdResponse(id=1630334, encodeId=8e98163033404, content=<a href='/topic/show?id=44e0e449217' target=_blank style='color:#2F92EE;'>#神经网络#</a>, beContent=null, objectType=article, channel=null, level=null, likeNumber=72, replyNumber=0, topicName=null, topicId=null, topicList=[TopicDto(id=74492, encryptionId=44e0e449217, topicName=神经网络)], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=e6ff21497204, createdName=by2016, createdTime=Sun Sep 06 01:26:00 CST 2015, time=2015-09-06, status=1, ipAttribution=), GetPortalCommentsPageByObjectIdResponse(id=36410, encodeId=32163641028, content=太难了,搞不懂, beContent=null, objectType=article, channel=null, level=null, likeNumber=37, replyNumber=0, topicName=null, topicId=null, topicList=[], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=96871613250, createdName=medcardio, createdTime=Fri Sep 04 15:39:00 CST 2015, time=2015-09-04, status=1, ipAttribution=)]
  2. [GetPortalCommentsPageByObjectIdResponse(id=1582921, encodeId=2792158292163, content=<a href='/topic/show?id=f6e1611840d' target=_blank style='color:#2F92EE;'>#机器#</a>, beContent=null, objectType=article, channel=null, level=null, likeNumber=0, replyNumber=0, topicName=null, topicId=null, topicList=[TopicDto(id=61184, encryptionId=f6e1611840d, topicName=机器)], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=13dc16973127, createdName=ms6279672939590805, createdTime=Sun Sep 06 01:26:00 CST 2015, time=2015-09-06, status=1, ipAttribution=), GetPortalCommentsPageByObjectIdResponse(id=1630334, encodeId=8e98163033404, content=<a href='/topic/show?id=44e0e449217' target=_blank style='color:#2F92EE;'>#神经网络#</a>, beContent=null, objectType=article, channel=null, level=null, likeNumber=72, replyNumber=0, topicName=null, topicId=null, topicList=[TopicDto(id=74492, encryptionId=44e0e449217, topicName=神经网络)], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=e6ff21497204, createdName=by2016, createdTime=Sun Sep 06 01:26:00 CST 2015, time=2015-09-06, status=1, ipAttribution=), GetPortalCommentsPageByObjectIdResponse(id=36410, encodeId=32163641028, content=太难了,搞不懂, beContent=null, objectType=article, channel=null, level=null, likeNumber=37, replyNumber=0, topicName=null, topicId=null, topicList=[], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=96871613250, createdName=medcardio, createdTime=Fri Sep 04 15:39:00 CST 2015, time=2015-09-04, status=1, ipAttribution=)]
  3. [GetPortalCommentsPageByObjectIdResponse(id=1582921, encodeId=2792158292163, content=<a href='/topic/show?id=f6e1611840d' target=_blank style='color:#2F92EE;'>#机器#</a>, beContent=null, objectType=article, channel=null, level=null, likeNumber=0, replyNumber=0, topicName=null, topicId=null, topicList=[TopicDto(id=61184, encryptionId=f6e1611840d, topicName=机器)], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=13dc16973127, createdName=ms6279672939590805, createdTime=Sun Sep 06 01:26:00 CST 2015, time=2015-09-06, status=1, ipAttribution=), GetPortalCommentsPageByObjectIdResponse(id=1630334, encodeId=8e98163033404, content=<a href='/topic/show?id=44e0e449217' target=_blank style='color:#2F92EE;'>#神经网络#</a>, beContent=null, objectType=article, channel=null, level=null, likeNumber=72, replyNumber=0, topicName=null, topicId=null, topicList=[TopicDto(id=74492, encryptionId=44e0e449217, topicName=神经网络)], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=e6ff21497204, createdName=by2016, createdTime=Sun Sep 06 01:26:00 CST 2015, time=2015-09-06, status=1, ipAttribution=), GetPortalCommentsPageByObjectIdResponse(id=36410, encodeId=32163641028, content=太难了,搞不懂, beContent=null, objectType=article, channel=null, level=null, likeNumber=37, replyNumber=0, topicName=null, topicId=null, topicList=[], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=96871613250, createdName=medcardio, createdTime=Fri Sep 04 15:39:00 CST 2015, time=2015-09-04, status=1, ipAttribution=)]
    2015-09-04 medcardio

    太难了,搞不懂

    0

Baidu
map
Baidu
map
Baidu
map