Re:Weight-Agnostic Neural Networks
by msoad on Thursday June 13, 2019 @12:07PM
(#20176197)
if a well designed architecture can perform better than learned weights of a general architecture, doesn't it make sense we design a system that can learn to architecture itself as it is learning the weights? Maybe that's the famous "Learning to learn" paper?
[ Parent
| on HN ]
Yes it does: https://arxiv.org/abs/1905.01392