Very interesting talk by @AnimaAnandkumar on key challenges to advance AI: 1) unsupervised learning with disentangled representations, 2) enhance robustness for real world app with recurrent feedback and composition and 3) adaptation and continual learninghttp://event.on24.com/wcc/r/2402282/C565DC8EF21A6476AF6E02C3E1C62FF7 …
-
I liked "Neural networks with recurrent feedback" as an example of how robustness can be enhanced with a predictive coding like architecture but the code does not seem to be available unfortunately: https://arxiv.org/abs/2007.09200
