如何确定卷积层和密集层内的参数数量？
目前，我正在研究我的第一个神经网络，使用卷积来尝试和测试这个想法。我的项目目前涉及一个非常简单的系统，该系统使用监督学习对正方形和三角形进行分类。
我很好奇的一件事是如何确定卷积层中滤波器的数量。人们似乎通常使用 32 到 512 之间的值，但我不确定我自己的模型使用什么数字。对于密集层也可以这样说。我见过人们添加了多个不同数量的密集层和卷积层，我想知道这到底意味着什么。
Currently, I am working on my first neural network using convolutions to play around and test with the idea. My project currently involves a very simple system that uses supervised learning to classify a square and triangle.
One thing I am curious about is how to determine the number of, say, filters in a Convolution Layer. It seems people typically use values between 32 and 512, and I haven't been sure on what numbers to use for my own model. Same can be said for the Dense layers as well. I've seen people add multiple Dense and Convolution layers with varying numbers and I would like to be informed as to what exactly it all means.
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）
这一切都取决于您拥有的数据集类型以及一些命中和试验权重以及训练期间添加的偏差，大多数情况下我们需要在 dropput 期间调整权重，总体而言没有特定的规则。
还要尝试过滤器的数量，不同类型的图像有不同的模式，调整这些参数将有助于模型捕获这些模式，多少层不固定[尝试不同的层和过滤器，研究人员已经创建了像 Resnet 和 Alexnet 这样的标准架构，请通过为了更好的理解]
It all depends on type of dataset you have and some hit and trial weights plus bias which added during training,mostly we need to adjust weight during dropput,overall there is no particular rule for this.
Also try number of filters, different types of images have different patterns,adjusting those parameters will help model to capture those patters,how many layers is not fixed [try different layers and filters, researcher's have created standard architecture like Resnet and Alexnet please go through it for better understanding]