Check out Koala 🐨: a new chatbot from BAIR researchers fine-tuned on dialogue that approaches ChatGPT quality!
Work led by
@haoliuhl
@Eric_Wallace_
@ArnavGudibande
and Xinyang Geng
Blog:
Demo:
BAIR is launching a Research Experience for Undergraduates (REU) this summer, aiming to work with exceptional students from HBCUs and PBIs on cutting-edge AI research and to foster discussions on career pathways in AI. See details & eligibility criteria at
When and why can feed-forward networks replace recurrent neural networks without a loss in performance? Find out the answer by reading the new BAIR blog!
Looking to hire top AI talent?
We've compiled a list of the brilliant
@berkeley_ai
Ph.D. Graduates of 2024 who are currently on the academic and industry job markets. (Thanks to our friends
@StanfordAILab
for the idea!)
Check it out here:
Sample-efficient multitask deep RL, on real robots, directly from pixels! The key is to enable agents to set their own goals and autonomously practice reaching them. Read more on the new BAIR blog post by
@Vitchyr
& Ashvin Nair!
BAIR Blog post on PEARL--a meta-RL algorithm that reduces meta-training time by 20-100X:
This work will also be presented at ICML this week (Wednesday: 12:15pm talk, 6:30pm poster) by Kate Rakelly & Aurick Zhou w/ Deirdre Quillen,
@chelseabfinn
,
@svlevine
Meta-learning can enable a robot to imitate a vision-based skill from a single human demonstration! New BAIR blog post from Tianhe Yu and
@chelseabfinn
on "One-Shot Imitation from Watching Videos"
"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!
#xai
Big congratulations to
@BerkeleyISchool
and
@Berkeley_EECS
faculty Niloufar Salehi on winning an
@NSF
grant to develop language translation for high-stakes settings!
Huge congratulations to BAIR alumni
@georgiagkioxari
and
@phillip_isola
for winning the prestigious 2021 Young Researcher Award at
@CVPR
! Previous winners include Berkeley alumni
@jon_barron
in 2020 and
@inkynumbers
in 2017 — proud of our trailblazers in Computer Vision!
How can we design neural networks in a principled way? New work from
@DeWeeseLab
@berkeley_ai
explores a theoretically-motivated paradigm for doing just that for dense networks! Read about it here:
Berkeley researchers
@ashishkr9311
and Jitendra Malik (collab. with
@zipengfu
@pathak2206
) have released their work equipping legged robots with an ability to adapt to unfamiliar terrain like sand, mud, hiking trails, grass and dirt.
@SCSatCMU
@facebookai
Happy New Year! Check out Ray, a distributed scheduler compatible with common deep learning frameworks (TensorFlow, PyTorch, MXNet, etc.) and designed with reinforcement learning in mind!
New reinforcement learning library rlpyt in pytorch thanks to Adam Stooke from
@berkeley_ai
(and previously intern with me at
@DeepMindAI
). There are a whole suite of RL algorithms implemented and framework for small and medium scale distributed training.
Wondering how to exploit the modelling capacity of generative models to perform data compression? Bit-Swap turns hierarchical VAEs into lossless compression schemes for high-dimensional data, like images. Check out this blogpost (incl. explanatory video)!
One last blog post to end a great year of research at BAIR! "Physical Adversarial Examples Against Deep Neural Networks" by Ivan Evtimov, Kevin Eykholt, Earlence Fernandes, and Bo Li et al.
New work led by Ilija Radosavovic, Tete Xiao and Bike Zhang
@berkeley_ai
on learning bipedal locomotion on real hardware (Digit
@agilityrobotics
) with model-free RL and sim-to-real!
In an article published by
@WIRED
, Mike Jordan
@UCBerkeley
, Daron Acemoglu
@MIT
and Glen Weyl
@MSFTResearch
assert that society would benefit far more by developing AI approaches that complement human intelligence, rather than trying to imitate it.
Should you plan ahead or learn by trial and error? Read
@Vitchyr
's BAIR blog post to see how temporal difference models (TDMs) combine the advantages of model-based and model-free reinforcement learning!
Laura Waller has been selected as the 2021 recipient of the Optical Society of America Adolph Lomb Medal. The Adolph Lomb Medal, established in 1940, is presented to an early career researcher who has made a noteworthy contribution to optics.
@optrickster
Vision systems require the ability to recognize common, rare, and never-before-seen classes in real-world scenarios. Check out the new BAIR blog on "Open Long-Tailed Recognition” (OLTR)!
Check out pics of commencement & our new BAIR PhDs! (Please add your own pics if you attended. Anything in the album can be reposted by anyone in BAIR!)
Congratulations to Prof. Stuart Russell on winning the
@IJCAIconf
'22 Award for Research Excellence, arguably the highest honor in the field of AI!
He joins a brilliant group of scientists including McCarthy, Minsky,
@geoffreyhinton
and fellow
@UCBerkeley
Profs. Malik and Jordan
The 100 most cited AI papers for 2022.
A detailed analysis of the most cited papers for the last three years allows good insights into the organisations and countries publishing the most impactful AI research right now.
Read here:
A thread 🧵
Robots that use tools! Through imitation and self-supervised learning, a robot can learn to make use of even unfamiliar objects as tools -- check out the new BAIR blog to find out how!
Check out this awesome project that BAIR student Greg Kahn () worked on at
@SkydioHQ
on training an autonomous deep neural network pilot to film while avoiding obstacles!
Skydio has completely reinvented the way drones fly, track subjects and avoid obstacles, but how did we get here? With lots of hard work and effort from the Skydio engineering team, that's how! Learn more about the engineering process in this week's Skydio Autonomy blog post.
Want to train your simulated characters to perform acrobatics and martial arts? Check out the new BAIR blog post from
@jasonpeng0
to learn how to combine deep RL with motion simulation from computer graphics!
If you're interested in the intersection of artificial intelligence and humanitarian assistance & disaster response (
#HADR
), please consider attending or submitting to the workshop on AI+HADR at
#NeurIPS2019
!
Curiosity provides an intrinsic reward function that generalizes across game levels! Check out this collaboration between BAIR and
@OpenAI
for an in-depth study of curiosity-driven learning on 54 environments, including Atari & Super Mario. 👾
We have released our paper on large-scale curiosity-driven learning. Our purely curious agents learn to play games, "walk", form rallies against each other across 54 benchmark environments without using any *external rewards*!
In collaboration with
@OpenAI
Tomorrow at 12 PM PT, the Berkeley AI lecture series continues.
@svlevine
will present "Reinforcement Learning with Large Datasets: A Path to Resourceful Autonomous Agents." Tune in live:
Interested in semi-autonomous robots and human-machine centaur teams? Read
@elbowproxy
’s new BAIR blog post on augmenting human control with deep reinforcement learning for flexible human-robot shared autonomy!
The BAIR REU is back!
@Berkeley_AI
is currently recruiting undergraduates from HBCUs and PBIs for a hands-on summer 2022 research experience in artificial intelligence. More details about eligibility and support at
Generative video models are rapidly improving in quality. Meet Replay, a new AI model that can generate stunning videos from text.
Replay v0.1 is designed to create ultrasmooth HD videos with a new interface. Available today for everyone.
What's New?
1. Replay understands plain…
BAIR faculty member Alexandre Bayen and BAIR Ph.D. student Pulkit Agrawal developed a computer vision-based system to improve medical care of people with Alzheimer’s
Check out the new BAIR blog post on AdaSearch, a new algorithm for adaptive sensing problems! AdaSearch shortens expensive data collection processes in spatial data acquisition problems like radioactive source-seeking via a successive elimination strategy.
We’re excited to announce the Berkeley AI Research Climate Initiative! The BCI aims to foster the development of fundamental AI research through directly working on impactful problems related to the most pressing issue of our time: climate change.
Congratulations to Professor Michael Jordan - in the past few months, he was elected as a Foreign Member of the
@royalsociety
, selected for the Vannevar Bush Faculty Fellowship by the
@DeptofDefense
, and awarded the 2021 Ulf Grenander Prize by the
@amermathsoc
! WOW!
Join us next Tuesday 12-1 PM (in-person + livestreamed) for a special panel on The Past and Future of Robotics and Machine Intelligence! Hear from 250+ years of combined experience w/ Ruzena Bacjsy,
@rodneyabrooks
,
@Ken_Goldberg
,
@JitendraMalikCV
, Shankar Sastry, Claire Tomlin.
@DeWeeseLab
Along the way, the authors show that a deep fully-connected net can be squashed into a shallow one with a special activation function! You can find the full paper at .
@DeWeeseLab
Sayan Gul (Berkeley undergrad in EECS/CogSci '18) died this summer of a heart attack en route to
#cogsci2018
. His friends, family & colleagues are raising funds to start a travel award in his name in support of undergraduate research in cognitive science.
Registration is now open for an exciting workshop organized by
@ask1729
and
@jlistgarten
at the
@SimonsInstitute
June 10th-14th in Berkeley, AI≡Science: Strengthening the Bond Between the Sciences and Artificial Intelligence.
📢 Excited to release Gorilla🦍 Gorilla picks from 1000s of APIs to complete user tasks, surpassing even GPT-4! LLMs need to interact with the world through APIs, and Gorilla teaches LLMs APIs. Presenting Gorilla-Spotlight demo🤩
Webpage:
The latest BAIR blog post from Tijana Zrnic discusses the reproducibility crisis in science and proposes a method for online false discovery rate (FDR) control, aiding reproducibility in large-scale statistical testing!
How can computers learn to follow instructions from a small amount of language data? Check out this BAIR blog post by David Gaddy on data-efficient language learning by pre-forming representations of an environment!
Congratulations to
@berkeley_ai
professor Claire Tomlin on her election to the
@theNAEng
for her contributions to safety-focused control of cyberphysical systems!
There’s been a lot of buzz about
#ChatGPT
on Twitter lately — the main architect behind the system is none other than
@berkeley_ai
alumni and OpenAI cofounder
@johnschulman2
! Hats off to John and the team!
Congratulations to BAIR faculty member Michael I. Jordan for receiving an honorary doctorate from Yale!
Yale's honorary degrees signal pioneering achievement in a field; past recipients include Benjamin Franklin, Marie Curie, & Stephen Hawking.
#Yale2020
Honorary Degrees!
Jim Allison, Medicine
Jerome Alan Cohen, Laws
Christiana Figueres, Humane Letters
Renée Fleming, Music
Herbie Hancock, Music
Michael I. Jordan, Engineering
Ming Cho Lee, Arts
Svante Pääbo, Science
Paula Vogel, Letters
BAIR faculty members join AI
@TheHouse
in bringing together academia, startups, industry, and education to build an ecosystem for AI, right in Berkeley!
Delayed outcomes should matter when evaluating "fair" decision-making procedures. Read the latest BAIR blog post on the delayed impact of fairness in machine learning to find out why!
BAIR@
#CVPR2019
: To see the work being presented by BAIR students and faculty please reference the 2019 Guide to Berkeley
@CVPR
: . The guide is a work in progress and is being updated.
We look forward to sharing the latest research from Berkeley with you!
Check out our 3 new BAIR blog posts in the last week on predicting visual affordances for robotics, turning deep neural nets into more interpretable wavelets, and open problems in ML safety!
Deep RL for real-world dextrous manipulation with a low-cost robot arm! The latest BAIR blog post by Henry Zhu, Abhishek Gupta, Vikash Kumar,
@aravindr93
, and
@svlevine
is online at
Check out Aviral Kumar and Avi Singh's
@avisingh599
new blog post, "Offline RL: How Conservative Algorithms Can Enable New Applications"
Also check out
@svlevine
& Aviral's NeurIPS 2020 tutorial on offline RL
Another new BAIR blog post: Sylvia Herbert, David Fridovich-Keil, Claire Tomlin and coauthors on "Ensuring Safe Real-Time Navigation of Dynamic Systems"
Our new work on redefining probabilistic models of human behavior for human-robot interaction has been accepted to
#HRI2020
! Take a look: . Joint work with Dexter Scobee, Jaime Fisac, Shankar Sastry, and
@ancadianadragan
.
Thought-provoking and timely lecture by Prof. Stuart Russell, co-author of the most popular AI textbook in the world, on "How Not to Destroy the World with AI":
New work from
@berkeley_ai
and
@CMU_Robotics
on visual locomotion enables a robot dog walking on tall bar stools in
@ashishkr9311
's living room -- entirely from onboard cameras and compute. Trained in simulation and deployed directly in the real world!
Check out our robot dog walking on stepping stones in my living room! Can also do stairs, construction sites, slippery slopes, etc -- all with just a single onboard RGBD camera, onboard compute, and no maps of the environment!
Meet the 2022
#CarnegieFellows
! Through humanities and social sciences research, these scholars are helping us better understand the past, the enduring challenges we face, and how we can create a better future for us all. Congratulations, fellows! 🎉
People give massive amounts of personal data to companies every day. Compensating users for their personal data requires an approach to assigning value to it. Ruoxi Jia's BAIR blog post presents a possible solution--a new method for data valuation.