Berkeley AI Research

Congratulations to David Gaddy and Dan Klein for the Best Paper Award at EMNLP 2020! Paper: https://www.aclweb.org/anthology/2020.emnlp-main.445/ Talk: https://slideslive.com/38939073 https://syncedreview.com//emnlp-2020-best-paper-award-goe/

It's thrilling to see how Ray has gone from a research project to production! Join us at #RaySummit to hear the latest in scalable ML and distributed systems. It starts tomorrow!

Congratulations to Shiry Ginosar for being selected CIFellow!

Congratulations to BAIR faculty Peter Bartlett, Bin Yu, Shankar Sastry, Yi Ma and @JacobSteinhardt, all appointed to @NSF Collaborations via their work through the @SimonsFdn! https://t.co/Y5VCVK4ZJv Berkeley AI Research (@berkeley_ai) September 4, 2020

"Explaining what Explainable AI Did Not": @lvinwan describes how to make models as accurate as neural networks but with an interpretable decision process in the new BAIR blog! https://t.co/nvJois1H8j #xai https://t.co/COqSPqQzgH (via Twitter)

RT @realNingYu: New paper on improving minority mode coverage of StyleGAN2: we combine GAN and IMLE objectives to get the best of both worlds. Joint work with @KL_Div , Peng Zhou, Jitendra Malik, Larry Davis, and Mario Fritz. More details in https://t.co/H2pOLGAWCy. Code and models coming soon. https://t.co/jAXt4LlQGN (via Twitter)

This NYT article on self-supervised learning also features BAIR faculty @pabbeel and @svlevine! https://t.co/YfTUhPas6g (via Twitter)

Congratulations, @andreaa7b & Dexter and co-authors, on being awarded best paper at @HRI_2020! https://t.co/WWu1o2db8G https://t.co/WA6NMzujmG (via Twitter)

Check out this BAIR work on detecting and exposing general artifacts in images generated by convolutional neural networks (CNNs)! Project website at https://t.co/qofq5ISYIh! https://t.co/Njq022wa09 (via Twitter)

The https://t.co/vjvHEGa7tr Digital Transformation Institute is a collaboration between https://t.co/vjvHEGa7tr, @Microsoft, and top universities to research #COVID19 and will be co-directed by BAIR professor Shankar Sastry. https://t.co/nVuBSzkIAa (via Twitter)

Representing scenes as neural radiance fields for view synthesis: A BAIR / Google research / UCSD collaboration! https://t.co/DOHZPoBYhi (via Twitter)

RT @shengs1123: Ever wondered why BN is not used in NLP? We found that NLP batch statistics exhibit *large variance* throughout training, which leads to poor BN performance. To address this, we propose Power Norm that achieves SOTA vs. LN/BN. [1/5] https://t.co/lpSJker4wV (via Twitter)

RT @svlevine: In deep RL, the training distr. plays a crucial role, and "standard" on-policy collection is *not* the best choice! It can produce non-convergence. Our method DisCor corrects this, substantially improving results. arxiv: https://t.co/9Cam6b89vo Blog: https://t.co/aU6EGeG3zs https://t.co/vtCHWBgU0J (via Twitter)

RT @NVIDIAEmbedded: Meet #BADGR: @berkeley_ai's autonomous self-supervised learning-based navigation system, running on NVIDIA's #JetsonTX2. BADGR can be trained with data gathered in the real-world, without any simulation or human supervision. https://t.co/HDfAauPlEF (via Twitter)

RT @svlevine: Greg Kahn's blog post describing BADGR, our real-world reinforcement learning system for robotic navigation that gets better and better with experience, is now out! https://t.co/aLQhR25AV3 https://t.co/nj6IfrIWz8 (via Twitter)

RT @Eric_Wallace_: Not everyone can afford to train huge neural models. So, we typically *reduce* model size to train/test faster. However, you should actually *increase* model size to speed up training and inference for transformers. Why? [1/6] https://t.co/GcjytCEmox https://t.co/HatYO5GfhP https://t.co/ivKyNo1ve0 (via Twitter)