Jian Sun
Person on 2 sites
Computer vision researcher, a co-author of residual networks. Led the visual computing group at Microsoft Research Asia and was chief scientist at Megvii.…
-
ppll (opens ppll.app)
Where public figures publish, and what they are on record saying.
Updated 2015-12-10 -
4 beliefs on korrents (opens korrents.com)
-
Depth is not free: past a point, deeper neural networks are harder to train rather than simply better. (opens korrents.com)
“Deeper neural networks are more difficult to train.”
-
A layer should learn a residual with reference to its own input rather than an unreferenced function, which is what makes great depth trainable. (opens korrents.com)
“We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions.”
-
Residual networks are easier to optimize than plain ones and keep gaining accuracy as depth increases. (opens korrents.com)
“We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth.”
-
Depth is not free: past a point, deeper neural networks are harder to train rather than simply better. (opens korrents.com)