Eric Jang: If DQN and GATO are overhyped (imply...| Robocrunch AI

If DQN and GATO are overhyped (implying somewhat obvious / easy to do?), can you suggest some similar will-be-overhyped ideas for the "deep learning junkies" to try? It's easy to trivialize other people's work in hindsight, was wondering if you could predict the future instead.
— Eric Jang (@ericjang11) May 13, 2022

GATO cites @lukaszkaiser 2017 ("one model to learn them all"), which had essentially the same high level idea and empirical findings: "kind of works, but not a slam dunk yet for the narrative of cross-domain generalization yet" https://t.co/nHmXyAioHk

Shared by Eric Jang at 5/13/2022

The same "implicit regularization" that makes it easy to train deep nets with supervised learning could make it hard to train them with RL! @aviral_kumar2 will present our analysis (DR3) at #ICLR2022 Wed 4/27 in a spotlight talk 10:30 am PT: https://t.co/td4DFQvvp8 https://t.co/vQ709H0GLh

Shared by Sergey Levine at 4/27/2022

The GATO paper buries this result a bit, but training on everything doesn't "just work" (yet). Zamir et al 2018 (Taskonomy) https://t.co/dRoG0Rs0PD suggests that in the presence of finite model capacity, whether some tasks positively or negatively transfer is non-trivial.

Shared by Eric Jang at 5/13/2022

In SF for the @reworkdl #DeepLearning Summit! I will talk at 2:30pm about enabling safely learning from experience at scale using self-supervised learning and analysis-by-synthesis! Some DL theory and a lot of cool Computer Vision results! Papers on arxiv: https://t.co/GGvfPar2J1 https://t.co/cuT4oegr9r

Shared by Adrien Gaidon at 2/17/2022

Why is deep learning so transformative and why is it worth trying for your project? (After you've tried something simpler first, that is.) https://t.co/3fA8QcH45b

Shared by Cassie Kozyrkov at 5/15/2022

Learning how to predict the future is hard—it doesn't have to be. See how NVIDIA deep learning experts use the open-source Time Series Prediction Platform to accelerate their GPUs: https://t.co/UyyQxXitFe #DataScience, #PyTorch, #Triton

Shared by NVIDIA AI at 2/16/2022

768