神经网络设计
我正在研究神经网络，想知道专家如何决定有多少个神经元就足够了，或者说知道什么范围是熟练的。是根据启发法做出决定还是根据应用程序有逻辑量？
举例来说，我想要一个简单的网络能够用 9x10 的网格来识别字母 I。
0 0 1 1 1 1 1 0 0
0 0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 0 0
0 0 1 1 1 1 1 0 0
所以这显然有 90 个输入神经元。为简单起见，假设有一个输出神经元（如果是字母 I，则为 0 或 1）。我的问题是，基于这个网络和应用程序设计，多少个隐藏层神经元足以产生更少的错误？人们是如何得出这样的结论的呢？
I am playing around with Neural Networks and wondered how the experts decided how many neurons is enough or say know what range would be proficient. Does one decide based on heuristics or are there logical amounts based on the application?
Say for example I wanted a simple network to be able to tell the letter I with say a grid of 9x10.
0 0 1 1 1 1 1 0 0
0 0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 0 0
0 0 1 1 1 1 1 0 0
So this would have obviously 90 input neurons. Let's say for simplicity, one output neuron (0 or 1 if it is the letter I or not). My question is, based on this network and application design, how many hidden layer neurons is enough to produce less error? How does one come to such a conclusion?
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论
评论（1）
这里有一个常见问题解答，看起来很有帮助。我认为没有任何已知的公式或任何具体的东西，它可能取决于输入域。
实际上对于这么简单的事情你不需要太多。
Here's a FAQ that looks helpful. I don't think there's any know formula or anything specific, it probably depends on the input domain.
Actually for something that simple you wouldn't need very much.