6 ms·
I find it strange you mention transfer learning, since one of the reasons neural networks are so popular are because they tend to excel at it. Adapting (i.e. fi
by relate 9y ago
I find it strange you mention transfer learning, since one of the reasons neural networks are so popular are because they tend to excel at it. Adapting (i.e. fine-tuning) networks trained on a task with a lot of data (e.g. image classification on ImageNet) to different tasks, such as image segmentation, has proven a very successful approach.
- taeric 9y agoThat was my intent in mentioning them. My understanding, though, is that we don't have a solid foundation on why it works. Just that that is the direction we are moving, and neural networks are good at it.