learning-notes
learning-notes copied to clipboard
daa233
如何理解 max-pooling 的平移不变性？
在读 Deep models for brain EM image segmentation: novel insights and improved performance 这篇论文时，看到文中写道：
The key challenge was to add as many convolution layers as possible without losing the translation invariance advantage that the max-pooling layers provide.
对于 max-pooling 的“translation invariance advantage”该怎么理解？
Max pooling achieves partial invariance to small translations because the max of a region depends only on the single largest element. If a small translation doesn’t bring in a new largest element at the edge of the pooling region and also doesn’t remove the largest element by taking it outside the pooling region, then the max doesn’t change. --- Ian Goodfellow from Quara https://www.quora.com/How-exactly-does-max-pooling-create-translation-invariance