You can get something like 5% error rate on MNIST with a well tuned linear classifier, it's not comparable with ImageNet. Note that the computer vision techniqu

You can get something like 5% error rate on MNIST with a well tuned linear classifier, it's not comparable with ImageNet. Note that the computer vision techniques used before convnets used things like SIFT features, which are another way of imposing a (sort-of) prior. I do believe that some sort of strong prior is necessary for the problem.