https://deepnote.com/@martin-molnar/Gradient-Pseudo-Swap-F2u_aTbqTOisNYCH_JORwg Martin MolnarGradient Pseudo-Swap “When we have layers in a neural network we want to train with gradient descent, but those layers don’t have smooth gradients that can be used,Continue readingWhat I Read: Gradient Pseudo-Swap

http://www.offconvex.org/2020/10/21/intrinsicLR/ Mismatches between Traditional Optimization Analyses and Modern Deep LearningZhiyuan Li and Sanjeev AroraOct 21, 2020 “You may remember our previous blog post showing that it is possible to doContinue readingWhat I Read: Mismatches between Optimization Analyses and Deep Learning