Pulkit Agrawal
@pulkitology

Professor @ MIT, Co-Founder SafelyYou Inc and AI Foundry.




Pulkit Agrawal    @pulkitology
What is 3D? It is mathematically well-defined. But what about in AI? We think we know, but do we? Turns out just like objects what we mean by 3D changes with the task. If curious, read the comment below. It shows that it's hard to talk about 3D representations without a task.

Pulkit Agrawal    @pulkitology
Roboticists are in two camps: either they make hardware or control algorithms. However, co-optimizing design and control can substantially improve performance! Our #RSS2021 paper proposes a fully differentiable method for such co-optimization: https://t.co/lgUdXdnYWq

Pulkit Agrawal    @pulkitology
Excited to announce that we are kicking off the summer edition of the CSL seminar series (https://t.co/pOcIRnAnyK) with a talk by @hausman_k on Multi-Task Robotic Reinforcement Learning at Scale. Join us at 11:30a EST on July 12.

Pulkit Agrawal    @pulkitology
Models reduce the need for data in decision-making but often pick up on spurious features. With Xiang, Ge, and Tommi Jaakkola, we propose TIA that relies on co-operative reconstruction for identifying important features in our #ICML2021 paper, https://t.co/mbOudztMcv

Pulkit Agrawal    @pulkitology
Try out the colab assignments https://t.co/lWp3m5vGPF Training sensorimotor learning systems is half science, half art!

Pulkit Agrawal    @pulkitology
Have a recent result in Self-Supervised learning? Consider submitting it to the ICML workshop: https://t.co/PvJ27J9omd It's due June 17th! #ICML2021 #AI Looking forward to seeing you and your work :)











 








Pulkit Agrawal

Sergey Levine

Facebook AI

Aran Komatsuzaki

hardmaru

Brenden Lake @ CogSci2021