Foundation Models & AI Agents for the Real World | AI Innovation from Inception to Societal Impact | @genesismolai • @GoogleDeepMind • @Waymo • @SandiaLabs
Palo Alto, CA
- Tweets 569
- Following 555
- Followers 3,414
- Likes 1,254
Pinned Tweet
I’m humbled to share that I’ve been elevated to IEEE Fellow, Class of 2026!
Citation: "For technical contributions to learning-based scalable autonomy and foundation models."
Research is a team sport. This belongs to the incredible mentors & collaborators who shaped my path.
ieee-org.widen.net/s/qqrqr2n…
5
25
5,230
I am hiring a research engineering lead for my team @GoogleAI. If you have background in autonomy, autonomous systems, and reinforcement learning, passion to make a difference, and are a true team player, apply directly:
careers.google.com/jobs/resu…
2
50
228
Very excited to share our recent work, which enables automated design of new RL algorithms.
We will be presenting it as oral at ICLR on Wed, 5-May.
sites.google.com/corp/view/e…
w/@JDCoReyes @yingjieMiao @dypeng @estebitius @svlevine @quocleix @honglaklee @AleksandraFaust
Today we present a new approach for automated discovery of generalizable #ReinforcementLearning algorithms that evolves a population of loss functions — represented by computational graphs — over a set of simple training environments. Learn more at goo.gle/3dFzDML
5
39
210
Bittersweet feelings today, as I wrap up nearly 10 incredible years at Google! Immensely grateful for the amazing colleagues, the friendships, and the chance to contribute to impactful projects like @Waymo, Google Brain Robotics/RL, Gemini Next Self-Improvement & Foundational Research at @GoogleDeepMind. What a journey!
While it's tough to close this chapter, I'm excited for what's next. Will be sharing more details soon. For now-- a massive thank you to everyone at Google! 🙏 Onwards!
#GooglerAlumni #AI #Robotics #ThankYouGoogle #NextChapter #Bittersweet
15
2
211
21,380
Personal update: I’m beyond excited to share that I have joined @genesistxai as Chief AI Officer!
It's an honor to join this mission and use #AIForGood to forge new treatments and deliver hope to families facing severe diseases. The GEMS platform is at the forefront of AI-driven drug discovery, and I'm incredibly proud to help lead the charge in this new era of Molecular ML. Looking forward to building the future with the amazing team at Genesis!
22
4
179
16,851
Join my team at @genesistxai ! 🧬 We're forging AI foundation models to unlock groundbreaking therapies for patients with severe diseases.
We're hiring ML Scientists, Engineers, TPMs & Interns in foundation models, #LLMs , #RL, #diffusion models, and other cutting-edge areas of #genAI.
Come solve challenges in #AI4Science! (Burlingame, NYC & Remote).
Apply: genesistherapeutics.ai/caree…
#Hiring #AI #MachineLearning #DrugDiscovery #Biotech #GenerativeAI #AI4Good
1
9
112
14,232
Very excited to share this work that with a simple idea reduces carbon footprint of training RL model by up to 3.8x w/ @Srivatsan91 , @profvjreddi, and other collaborators from @hseas and @DeepMind.
Introducing ActorQ, a novel paradigm that uses a quantization optimization technique to speed up #ReinforcementLearning training and reduce RL training’s carbon footprint, while maintaining performance. Learn more at → goo.gle/3RgCEUC
1
11
85
A little sneak peak, #CoRL2021 is moving to #OpenReview. Submission deadline on 18-June. Official info coming up soon @corl_conf.
openreview.net/group?id=robo…
1
29
81
What's the best way to learn deep RL? Write a book, of course! Congratulations @lgraesser3! Honored to work with you!
Foundations of Deep Reinforcement Learning: Theory and Practice in Python (Addison-Wesley Data & Analytics Series) amazon.com/dp/0135172381/ref…
2
9
65
Can we learn RL algorithms that outperform existing ones, generalize to a wide variety of environments, and can be analyzed?
Evolving Reinforcement Learning Algorithms
Friday @ Deep RL Workshop at @NeurIPSConf
Room: B.b6, 12:30-13:30 and 18:00-19:00 PT.
4
4
58
Deeply honored to receive Early Government or Industry Career Award in Robotics and Automation @ieeeras to be presented @icra2020. Many thanks to all of my collaborators and mentors without whom this would not be possible.
Congrats to the other winners!
ieee-ras.org/about-ras/lates…
6
2
58
Multi-Agent Reinforcement Learning for Microprocessor Design Space Exploration, Krishnan et al., Oral #MLSyS, arxiv.org/abs/2211.16385
Hardware architecture design is a search over a complex interconnected system. #MARL brings up 60x improvement over single RL agents.
1
12
41
It was a great honor to give a keynote at @iros2022 and share some of the team's recent work toward generalist agents in real world and scalable autonomy.
Grab your seats fast! The Keynote sessions have started. We have Vivian Chu (@dr_vchu),@DiligentRobots as our very first speaker. Followed by Aleksandra Faust (@AleksandraFaust), Google Research, and James Kuffner (@james_kuffner), @ToyotaMotorCorp #iros2022
2
38
Excited to attend @icmlconf in Vienna. I'll be here the whole week. Let me know if you want to catch up. I am looking forward to connecting with everyone. Here is some of the work to highlight.
1
6
38
7,519
Another @corl_conf sneak peek....
In addition to the regular research papers, #CoRL2021 is soliciting Blue Sky Idea positional papers.
What is a big, crazy idea that we should be doing? Does robot learning have wrong assumptions? Do tell. :)
Also due on 18-June.
7
36
Our #AutoRL reward search benchmarking paper will appear at AutoML@ICML. You can read it at: arxiv.org/abs/1905.07628
1) More gain on difficult tasks.
2) Under the fixed training budget, reward shaping is more effective.
3) Simple objective is viable option.
w/ @xenotaur
14
34
Common HTML understanding tasks can be done without custom NN architecture design and with orders of magnitude less data by fine-tuning LLMs. Bidirectional attention appears to be crucial, and context windows remain the bottleneck.
"Understanding HTML with Large Language Models" Our newest work shows that LLMs pretrained on standard text corpora transfer remarkably well to web-based tasks. We achieve a new SOTA on supervised MiniWoB: 50% better perf with 200x less data than prev best arxiv.org/abs/2210.03945
1
6
31
Putting a limit of number of pages for references takes away the authors' focus and energy away from what matters most, the content of papers, and disadvantages papers with lengthy citations. So, we fixed it. Use the precious time on finding better insights.
#CoRL2021 now has unlimited references for papers!
June 18th is the submission date for regular papers - get them in soon. Find out more at: robot-learning.org
#robots #learning #machinelearning #robotics #conference #robot #publishing
1
30
Very excited to see Waymax launch and looking forward to the innovation that it might bring to the autonomous driving space. Last but not least, it was a wonderful collaboration between #waymoresearch and @GoogleDeepMind w/ @georgejtucker @JDCoReyes @BeccaRoelofs @agarwl_
Today, we’re introducing Waymax, a first-of-its-kind simulator developed specifically for solving autonomous driving research problems around planning and sim agents. Discover how researchers can access Waymax: waymo.com/blog/2023/10/wayma…
9
27
5,426
Some really cool work from my team. Best paper candidate @iros2022
Our paper on representation learning for learning better *real world* robot policies was just posted on arxiv. This will be presented at IROS 2022 as a Best Paper Award Finalist! (anyone else going to Kyoto??) @kuanghueilee arxiv.org/abs/2207.13224
piped.video/watch?v=59vMC3fT…
25
It is very exciting to have a RL benchmark verified in the real world! I am looking forward to seeing how it pushes the boundary in RL research.
The BLE blog post is out today! If you've been curious to find out more about how the simulator works or why this is a great challenge for RL methods – take a look!
1
2
24
Our new preprint introduces a genestlist web agent. Txt->multi-step actions (code) on the websites in the wild. Three #LMMs: planner, perception, and controller with self-supervision
arxiv.org/abs/2307.12856
@frt03_ @IzzeddinGur @austinvhuang @mustafa_safdari @douglas_eck
A Real-World WebAgent with Planning, Long Context Understanding, and Program Synthesis
paper page: huggingface.co/papers/2307.1…
Pre-trained large language models (LLMs) have recently achieved better generalization and sample efficiency in autonomous web navigation. However, the performance on real-world websites has still suffered from (1) open domainness, (2) limited context length, and (3) lack of inductive bias on HTML. We introduce WebAgent, an LLM-driven agent that can complete the tasks on real websites following natural language instructions. WebAgent plans ahead by decomposing instructions into canonical sub-instructions, summarizes long HTML documents into task-relevant snippets, and acts on websites via generated Python programs from those. We design WebAgent with Flan-U-PaLM, for grounded code generation, and HTML-T5, new pre-trained LLMs for long HTML documents using local and global attention mechanisms and a mixture of long-span denoising objectives, for planning and summarization. We empirically demonstrate that our recipe improves the success on a real website by over 50%, and that HTML-T5 is the best model to solve HTML-based tasks; achieving 14.9% higher success rate than prior SoTA on the MiniWoB web navigation benchmark and better accuracy on offline task planning evaluation.
2
27
3,129
Very excited to see the Web Agents line of work land in Bard Extensions as part of building @youtube and @google extensions. Congrats @austinvhuang and @yujin_tang and everyone else.
We’re adding extensions to Bard so you can connect it to your favorite Google apps including Gmail, Drive + Docs for even deeper collaboration. We’re also updating how we validate the claims in Bard’s responses with an improved “Google It” button + more.
blog.google/products/bard/go…
5
24
5,780
Really excited to see this paper out. It aims to start a practical conversation around model capabilities and risks associated with the capabilities and the context in which they are used. Huge thanks to @merrierm and my other colleagues from @GoogleDeepMind
Excited to share a new article I wrote with colleagues from DeepMind - "Levels of AGI: Operationalizing Progress on the Path to AGI" - arxiv.org/abs/2311.02462
5
24
6,856
Apparently, we are now Google Deepmind. I'm excited about this!
The phenomenal teams from Google Research’s Brain and @DeepMind have made many of the seminal research advances that underpin modern AI, from Deep RL to Transformers. Now we’re joining forces as a single unit, Google DeepMind, which I’m thrilled to lead! dpmd.ai/announcing-google-de…
1
23
4,572
Fundamental difference between offline RL and SL, helps shed the light if offline RL can be ever reduced to SL.
Our newest paper arxiv.org/abs/2112.12320 asks if there exists an analog to “validation error” (from sup learning) in an offline bandits/RL context. We identify 3 properties that any model selection algo should achieve, and indeed are achievable in both SL and online bandits/RL...
2
22
Thrilled to have presented work on autonomous agents through evolution, reinforcement, and self-supervision to over 3000 undergraduate students at @SforAiDL event this weekend in India.
This is exciting time for broadening the participation and outreach in #AI and #DeepLearning.
@AleksandraFaust's talk is live! Check it out at: crowdcast.io/e/summer-sympos…
or our live stream on YouTube: piped.video/channel/UCU5Pl1R…
1
4
24
It's been a blast to share part of your journey and work with you @agarwl_ ! Now, go do more amazing things. :)
1
21
5,003
Excited about upcoming #iros2021 and speaking at four workshops:
1. Self-supervision in Motion Planning & panel @ CLAMP
2. Learning to Learn for RL & panel @ SPAR
3. Panel on RL & Control & HCI @ RL-Conform
4. Autonomous navigation @ IPLC
2
1
21
Lab-grown black hole -- terrifyingly exiting.
livescience.com/black-hole-a…
1
5
20
There's always insecure people. Allowing them to instill self doubt, empowers them. Don't. Your work is outstanding. Your energy is precious. Carry on.
To anyone who feels unfairly supported or advantaged, remember, that's what a leveled playing field looks like. You deserve it
Pedro’s comments are so damaging. I’m applying for faculty jobs now and it feels like even if I get a good position, whether I *deserve* to hold that position will always be in question. I will continually have to re-prove that I am as qualified as my male colleagues.
1
20
Environment Generation for Zero-Shot Compositional Reinforcement Learning @ NeurIps, Poster session 1 now until 10:00 am PST
w/ @IzzeddinGur @natashajaques @yingjieMiao @honglaklee
openreview.net/forum?id=CeBy…
1
2
21
How do we design "just-the-right-challenge" curriculum and train navigation agents that generalize to complex unseen environments?
Adversarial Environment Generation for Learning to Navigate the Web
Fri, Deep RL Workshop @NeurIPSConf
Room: D.d6, 12:30-13:30 & 18:00-19:00 PT
1
3
21
Observe and learn to anticipate behaviors and motion patterns with self-supervision. Then use it to plan cooperative tasks.
One of my favorite papers from last year
w/ @rose_e_wang, J. Chase Kew, Dennis Lee, Tsang-Wei Lee, Tingnan Zhang, @brian_ichter, Jie Tan, @corl_conf 2020
Introducing a model-based #RL approach for robot navigation, called hierarchical predictive planning (HPP), that enables agents to align their goals on the fly in order to solve the decentralized rendezvous task. Learn more at goo.gle/3sZSyqb
5
20
Imitation Is Not Enough: Robustifying Imitation with
RL for Challenging Driving Scenarios, Y. Lu et al., #ML4AD, tinyurl.com/mrxwdssm
BC-SAC, trained 100k miles of urban driving data, substantially improves safety and reliability without compromising on human-like behavior.
1
7
18
Congrats to the authors of the accepted submissions:
openreview.net/group?id=robo…
Huge thanks to the ACs and reviewers who made a push to accelerate the decisions ahead of plan!
#CoRL paper notifications are out! There were:
400 main track submissions, 156 accepted papers inc. 26 orals.
Each submission had >=3 reviews.
53 area chairs, 350 reviewers, 1200 reviews + lots of discussion over #OpenReview improving papers. 38.25% acceptance rate
#CoRL2021
20
It was an honor and blast to speak about "Autonomous Agents in the Era of Large Language Models" and EEML 2024. Thank you for organizing the amazing event.
Slides: tinyurl.com/aallmeeml
EEML'24 Day 2 videos are out! 🇷🇸
* Bayesian DL (@yeewhye): piped.video/watch?v=rgEXgdEc…
* Intro to RL (Hamza Merzić): piped.video/watch?v=LcZ4vAqz…
* Autonomous Agents (@AleksandraFaust): piped.video/watch?v=GGAclFbg…
* RL Tutorial (@andreeadeac22 & Ognjen Milinković): piped.video/watch?v=nCKXzXrd…
2
3
20
2,345
ICLR 2026 in Rio!
Announcing the ICLR 2026 Call for Papers!
Abstract submission: Sept 19 (AoE)
Paper submission: Sept 24 (AoE)
Reviews released: Nov 11
Author/Reviewer discussion: Nov 11-Dec 3
Final decisions: Jan 22 2026
iclr.cc/Conferences/2026/Cal…
1
19
1,219
Excited about Learning in the Loop Systems workshop @icra2019. Friday, 24-May. Keynotes:
Marc Bellemare, @marcgbellemare @GoogleAI
Davide Scaramuzza, @davsca1 @ETH_en
Lydia Tapia, @LydiaETapia @UNM
Ashish Kapoor, @MSFTResearch
Gareth Cross, @SkydioHQ
uav-learning-icra.github.io/…
1
9
19
Exciting to see more work on learned curriculum and emerging complexity.
Evolving Curricula with Regret-Based Environment Design
Website: accelagent.github.io
Paper: arxiv.org/abs/2203.01302
TL;DR: We introduce a new open-ended RL algorithm that produces complex levels and a robust agent that can solve them (e.g. below).
Highlights ⬇️! [1/N]
1
1
15
Wednesday -- Stop Regressing, an oral, that shows that to scale deep #RL for large capacity models cross entropy classification loss should replace squared Bellman errors. Simple with comprehensive analysis and empirical results. @JesseFarebro's internship project.
Poster: 24 Jul 1:30 — 3 p.m. CEST. Hall C 4-9 #1311
Oral: 4A Reinforcement Learning 2
24 Jul 4:30 — 5:30 p.m. CEST
icml.cc/virtual/2024/poster/…
@JesseFarebro , @pcastr , and I will be there. w/ @agarwl_ and @aviral_kumar2 and others, who won't be able to make it.
1
4
16
1,006
100%, Kevin. The main thing teenagers in highschool need to come out with is an understanding of who they are and where their passions and values are. The only way to do that is with a healthy amount of unstructured time, experimentation, and yes, failures.
12
1,891
Remote talks take away a lot of the anxiety around giving talks at conferences @iclr_conf
1. Pre-recorded talks -- no stage fright.
2. Questions asked while videos playing back -- authors have time to think about the answers.
3. Sessions chairs -- talks are done, no no-shows.
1
1
15
Check out @GoogleAI blog post from @natashajaques and @MichaelD1729 on generative environment design for curriculum learning. What's cool is that the meta-learner makes no assumptions about the agent under training, opening possibilities of training other agents, including human.
Excited that our blog post about PAIRED is finally out! Joint work with @MichaelD1729, @EugeneVinitsky, @svlevine, @IzzeddinGur, @AleksandraFaust, and others.
1
13
Another posting just went live. My team is looking for a research SWE with experience in RL or LLMs, and interest in scaling up research and real world RL applications. w/ @natashajaques @ofirnachum @georgejtucker @BeccaRoelofs
tinyurl.com/rSWEBrainRL
1
15
Friday -- Many-Shot In-Context Learning, oral at Long Context Foundation Models workshop, which among other things, introduces unsupervised and reinforced in context learning.
26-Jul, Hall A2
Oral: 2:45-3:00O
Poster: 3:00-4:00
openreview.net/forum?id=8ul3…
Led by @agarwl_ and Avi Singh. Lei M. Zhang will present. w/ Bernd Bohnet, Luis Rosias, Stephanie Chan, Biao Zhang, Ankesh Anand, Zaheer Abbas, Azade Nova, @JDCoReyes , Eric Chu, @FeryalMP , @hugo_larochelle
1
4
16
3,290
Evolving multi objective policy gradient reinforcement learning loss functions that are performing, generalizing well, and are stable.
I'm excited to share MetaPG, the result of my work at @GoogleAI!
Will be presenting this work at the Generalizable Policy Learning workshop at #ICLR2022 next Friday 4/29 arxiv.org/abs/2204.04292
w/ @yingjieMiao, JD Co-Reyes, Aaron Parisi, Jie Tan, Esteban Real, @AleksandraFaust
14
From the top of my mind: Maja Matarić, Stefanie Forrest, Manuela Veloso, Maria Gini, Raia Hadsell, Danica Kragić, Leslie Kaelbling, Doina Precup
1
14
1,361
Database of the learned algorithms: github.com/jcoreyes/evolving…
3
13
Thanks #corl2022 for the amazing conference!
It was a privilege to give a keynote on real-world autonomous agents. I discussed our current and emerging work on the generalist agents, including the roles of model capacity and data quality: piped.video/watch?v=cdVKs3fU…
But wait, this is not everything!
We now have ALL OUR TALKS ONLINE!
This includes keynotes, tutorials, orals, spotlights, sponsors, opening & closing sessions...
corl2022.org/videos
13
12,171
Indeed. Very sad day. Samy is a person of integrity and of open mind. It certainly has been a privilege working with him.
Today is a very sad day for many of us. It has been a great privilege working with Samy. He is a real role model in research, management and being a true ally. Thank you, Samy!
13
Looking forward to presenting, attending, connecting with old friends, and making some new ones. Thanks for organizing.
We are beyond excited to announce the location for EEML2024: beautiful Novi Sad, Serbia, on the shore of the Danube, just after the EXIT festival and before ICML, 15-20 July 2024. Amazing speakers confirmed already, this is bound to be an epic experience! eeml.eu/
1
2
15
4,400
Also for cases where human examples are difficult (problem solving) unsupervised and reinforced in-context learning brings CoT without exemplars with similar or stronger performance. Plus, many shot ICT can correct for pre training biases, it just needs the right context.
We studied In-Context learning with hundreds to thousands of examples.
My favorite example: I sent *one million* tokens to Gemini 1.5 Pro for linear classification with 64 dimensional integer-valued vectors and many-shot learning performs similarly to k-Nearest Neighbours.
1
7
15
3,259
Thanks to all collaborators, especially @jparkerholder who pulled it together. This has been a great collaborative effort and hopefully it is useful to the community.
New Article: "Automated Reinforcement Learning (AutoRL): A Survey and Open Problems" by Parker-Holder, Rajan, Song, Biedenkapp, Miao, Eimer, Zhang, Nguyen, Calandra, Faust, Hutter and Lindauer jair.org/index.php/jair/arti…
9
It was an honor to be part of Quantum and AI panel and present about the future of AI and autonomy to the Space Studies, and Aeronautics and Space Engineering Board @theNASEM
nationalacademies.org/event/…
11
1,434
Nice interview with @nikiparmar09
medium.com/@spsayakpaul/an-i…
1
12
New work on interaction with quadrupeds. LLMs output foot patterns, which a separately learned controller executes. It's fascinating that LMMs decode non-descriptive motion prompts ("We are going to a picnic") into emotionally expressive motion (leaping, happy dog) @yujin_tang
SayTap: Language to Quadrupedal Locomotion
We use foot contact patterns as interface to bridge instructions in NL and low-level control commands.
New paper w/ Wenhao Yu, Jie Tan, @heiga_zen, @AleksandraFaust, @ttyharada
Web saytap.github.io
PDF arxiv.org/abs/2306.07580
1
12
1,177
With 8 hours to go, here is a process and criterion for desk rejects @corl_conf Double check the anonymity, formatting, and relevance to robot learning. If there is doubt, it goes for a full review.
Happy paper writing!
robot-learning.org/author-in…
2
11
Thank you for organizing. The discussion was engaging and fun. Specialist or generalis? Time will tell.
We had a really great panel discussion today at the EGG workshop at #RSS2023 with @jeffclune @AleksandraFaust @_rockt @mark_riedl @pathak2206!
Thank you to the speakers, the attendees, and my co-organizers for making today's workshop interesting and valuable!
1
12
1,183
Tuesday -- Levels of AGI, position spotlight paper in which we propose performance based definition, various use modalities, and risks of AI systems to operationalize progress on the Path to AGI. Let's discuss!
23 Jul 11:30 am — 1 pm CEST Hall C 4-9 #2306
icml.cc/virtual/2024/poster/…
w/ @MerrieRM,@JaschaSD, Noah Fiedel,@TrisWarkentin, @AllanDafoe, @Clmt, @ShaneLegg
6
1
9
779
Excited about mention of our work at #GoogleIO2021 and very proud of the team. #assisstivebots
"Assistant takes over the tedious parts of web browsing: scrolling, clicking and filling forms, and allows you to focus on what's important to you"
zdnet.com/article/google-io-…
10
Excellent summary on how to review papers. I especially like the tip about judging forward-looking approach to assessing impact -- impact to the future research.
The #NeurIPS2021 review period is starting soon, and many of us end up complaining about the quality of reviews.
So I decided to write a blog post describing how I approach reviewing, in the hope that it helps others.
Let me know your comments!
psc-g.github.io/posts/mentor…
7
What a scientist looks like to children changed over last 40 years. In 70s only 1% of children drew scientists as women regardless of child's gender. Today, more than 30% do, and more than half girls now draw scientists as women. #InternationalWomenDay2020
sciencemag.org/news/2018/03/…
10
I am heartbroken over the unfolding tragedy in Turkey and Syria. My thoughts are with those affected and thier families, and everyone on the ground. #TurkeyEarthquake
10
1,619
At @icmlconf? Stop by and learn about:
- Navigation in Dynamic Environments: Wed 12:30-2pm at @Google's booth.
- #AutoRL - Fri 9:30-11 am @ #AutoML workshop, arxiv.org/abs/1905.07628
- #SafeRL - Fri 10-11 am @ RL4RL, openreview.net/forum?id=SJgU…
3
10
Proud of team's presence at #ICRA2019. 3 papers and 1 workshop with @lewisprometheus @brian_ichter and @xenotaur
Details below.
1
1
9
@NeurIPSConf workshop time! Check out our papers:
"Imitation Is Not Enough...," #ML4AD w/ @Waymo on combining imitation learning and RL.
"MARL for Microprocessor Design...," #MLSyS w/ @hseas leads to <60x DRAM improvement (oral).
Plus 3 workshops my team is co-organizing.
1
9
Yup. Papers that were assigned to me were completely outside of my area. Worse, there was no way to contact AC.
In robotics, it is the responsibility of the AC/AE to find qualified reviews for a specific paper, and to ensure the quality of the reviews.
1
8
Finally, on Saturday, I am helping organize #AutoRL workshop, with an exciting list of invited speakers: @pcastr, @chelseabfinn, @jparkerholder, Pierluca D'Oro, and Roberta Raileanu.
27 Jul, Stolz 0
autorlworkshop.github.io/
1
8
357
To broaden the participation in the review process, we are opening up a call for both regular and reproducibility reviewers for this year's #AUTOML23. If you are interested, please submit your application below. You'll be notified if accepted.
Ensuring the highest quality standards is crucial for the success of any conference, including #AUTOML23. Therefore, we're currently seeking reviewers. If you have experience in #AutoML, your help would be greatly appreciated in making #AUTOML23 a success. docs.google.com/forms/d/1kJB…
1
7
2,439
Also, from earlier this year.
Neural Collision Clearance Estimator for Batched Motion Planning
w/ J. Chase Kew, @brian_ichter, @MaryamBandit, Tsang-Wei Edward Lee
#wafr2020
robotics.cs.rutgers.edu/wafr…
1
8
Huge congrats to @austinvhuang and the team!
I'm happy to share the release of gemma.cpp - a lightweight, standalone C++ inference engine for Google's Gemma models:
github.com/google/gemma.cpp
Have to say, it’s one of the best project experiences of my career.
6
730
@GeorgiaChal, thanks for asking. Deadlines are exhausting. However, @corl_conf has a very tight review schedule -- 3 months from submissions to notifications, including revisions! Sadly, we can't extend the deadline without jeopardizing the quality of the reviews or the timeline.
1
7
Second part looks at the generalist agents from the data perspective. From the imitation generalist agents that perform general tasks but are limited by the expert quality, to new RL foundational models, which outperform the data they are trained on, and scale with model size.
1
1
4
334
Happy to see RL, robotics, and system architecture come together to enable easier and safer training. Thanks @Srivatsan91 et al.
Full paper @MLJ_Social:
link.springer.com/article/10…
#Air Learning: A gym environment to train deep reinforcement algorithms for aerial robot navigation techxplore.com/news/2021-08-…
5
An inspiring group, and very happy to see some familiar faces. Congrats, @sarahtangy, Hana Lodge, and Shir Yehoshua.
businessinsider.com/rising-s…
7
Thanks for recognizing the work. For the record, the open sourcing is under way, but not yet available.
We tried image based methods, which didn't improve the performance. That led us to go with text representation and bypass the renderer all together.
3
7
Congrats to @Srivatsan91 and co-authors -- our paper "The Sky Is Not the Limit: A Visual Performance Model for Cyber-Physical Co-Design in Autonomous Machines" was selected a Best of IEEE Computer Architecture Letters in 2020.
ieeexplore.ieee.org/abstract…
1
1
7
Our Genesis Therapeutics ML team is headed to @icmlconf! I'll be there tomorrow. @alshedivat and David Li-Bland will be participating in the @genbio_workshop to engage in timely discussions on the acceleration of AI for biological research and drug discovery. If you’re attending, be sure to stop by.
1
3
9
840
Three exciting papers tomorrow (Thursday) at #ICRA2021 on navigation and motion planning.
Vision and Perception: Autonomous Vehicle Navigation III − ThKT1
Micro/Nano Robotics III − ThKT7
Probabilistic Method in Motion Planning − ThKT14
1
6
Excited to present some of our latest research on autonomous navigation at scale at Perception, Learning, and Control for Autonomous Agile Vehicles workshop at IROS. The talk starts in half an hour and will be recorded.
wp.nyu.edu/workshopiros2020p…
6
Thanks for writing this, Nathan. It is very insightful and very glad that you highlighted the networking as a long term benefit of the job search. Right after PhD is probably the only time one does a massive job search. It's time well spent that pays dividends. Congrats again!
1
5
Congrats, Adrian and @FootballPaly!
GSF Week #9 weightsandbars.com
Special Teams Player of the Week
@FootballPaly #50 Adrian Faust - DT
🙏🏽 Thank you all for supporting and following @gsfUNLIMITED @GetSportsFocus
5
Thanks Guillem for hosting and the audience for great questions.
🤖🌍 It is now available online the @AIforGood keynote where we explored the future of #RoboticsforGood with Vincent Vanhoucke, Senior Director of Robotics at @Google, in a session moderated by @AleksandraFaust
🌐 Watch it on piped.video/watch?v=fPEChRFi…
@ITU @ITUstandards
1
5
1,230
Check out #AirLearning, RL on edge research platform. Extends #AirSim and @OpenAI Baselines with Hardware-in-the-Loop (#HIL) and adds an environment generator for domain randomization.
w/ @Srivatsan91 @zaaddan
github.com/harvard-edge/airl…
1
2
6
My college days were marked with months of protesting a sitting president who refused to accept election results and leave the office: en.m.wikipedia.org/wiki/1996…
6
1k member on @united and eligible for a free upgrade. Four empty seats in business. The system won't let upgrade me. The gate agent couldn't do it, an agent on the phone couldn't do it, the pilot couldn't do it. Ten other people got upgraded. No explanation why. What the heck.
1
6
2,721
Friday at #RL4RL at #icml2019 workshop in Seaside Ballroom:
Lyapunov-based Safe Policy Optimization for Continuous Control
openreview.net/forum?id=SJgU…
Come to learn about RL algorithms that maintain safety constraints during training.
Poster times: 9-10am, 12:00-12:30pm.
1
3
6
Very exciting work!
Very excited to share a line of work on documentation for applied RL systems. Introducing,
Reward Reports for Reinforcement Learning
Paper: arxiv.org/abs/2204.10817
Template: github.com/RewardReports/rew…
Blog Post: bair.berkeley.edu/blog/2022/…
Workshop June 11: rewardreports.github.io/work…
🧵
1
5