The decision reasoning is super helpful to put things in context. The arbitrary binary decision to "accept" vs "reject" especially for the snooty "high bar for acceptance at ICLR" is laughable in a world of free information access.
If backprop is not needed, would this finding make automatic-differentiation functionality obsolete in DL frameworks, allowing these frameworks to become much simpler? Or is there still some constant factor that makes backprop favorable?
Reading the openreview link, the current understanding is that this approach is dramatically more computationally intensive than standard backprop - limiting its utility.
10 comments
[ 3.1 ms ] story [ 62.5 ms ] threadHe mostly points to this post in LessWrong - https://www.lesswrong.com/posts/JZZENevaLzLLeC3zn/predictive...
https://arxiv.org/abs/2103.04689