Ask HN: Why aren't convolutional decision trees or convolutional SVMs a thing?

2 pointsphenkdo1 comment

Put another way, what made ANNs more suited to convolutional feature extraction from imagery (vs. other ML methods)?

Decision trees and SVMs are too specialized for classification, which makes them a bad fit for feature extraction. ANNs can be stacked in layers and trained as a whole, but if you stack two decision trees, how do you propagate the error signal between them when the classification is incorrect?

On the other hand, you can use ANNs as the feature extraction stage for decision trees and SVMs, and there are people who do that.

Reply on news.ycombinator.com