Orr Krupnik (@orrkrup) header
Orr Krupnik (@orrkrup) profile picture

AI Researcher | Reinforcement Learning, Robotics | PhD Technion | More data isn't all we need 🦾 🌍📈

Israel
Joined April 2018
Photo by Orr Krupnik (@orrkrup)Photo by Orr Krupnik (@orrkrup)Photo by Orr Krupnik (@orrkrup)Photo by Orr Krupnik (@orrkrup)Photo by Orr Krupnik (@orrkrup)Photo by Orr Krupnik (@orrkrup)Photo by Orr Krupnik (@orrkrup)Photo by Orr Krupnik (@orrkrup)Photo by Orr Krupnik (@orrkrup)Photo by Orr Krupnik (@orrkrup)
Orr Krupnik retweeted
Incredibly proud to have featured as a partner in this release, securing Granite at the weight level without collateral damage in the model's core capabilities. Congrats to the entire @IBMResearch team, looking forward to continuing to push together!
Meet Granite 4.2, IBM’s latest family of open models purpose-built for enterprise agentic AI. With new native reasoning capabilities, Granite 4.2 can plan, reason, self-correct and reliably use tools to help automate complex enterprise workflows ⬇️
2
3
95
Orr Krupnik retweeted
Shelly will present this tomorrow (Tue) at 10:30am, visit our poster to learn how to SAFELY use VLAs! #ICML2026
Excited to announce our paper is accepted to #ICML2026! VLA robot policies are capable, but do they know when they're failing? We tackle this with TDQC: a framework for sequential uncertainty calibration, with a surprising link between calibration theory and RL value function.
3
13
1,275
Orr Krupnik retweeted
Excited to announce that our new book Reinforcement Learning Foundations is available for preorder! W/ @shiemannor and @YishayMansour 30% discount and free shipping till July 21 Discount code 148294 cambridge.org/il/universityp…
3
44
279
15,139
Orr Krupnik retweeted
VLA is 95% certain about current action. Will it 95% succeed in the task? Obviously, not necessarily. But if you’re clever, you can *calibrate* action prob. to task success. Our #ICML2026 paper formulates this + SOTA algorithms based on new connection to RL temporal differences
Excited to announce our paper is accepted to #ICML2026! VLA robot policies are capable, but do they know when they're failing? We tackle this with TDQC: a framework for sequential uncertainty calibration, with a surprising link between calibration theory and RL value function.
9
47
8,029
We're partnering with Tzafon, an AI R&D lab, to build next-generation agentic machine intelligence! Through our partnership, we will provide compute capacity & cloud services to train Tzafon’s new multi-agent models & develop new automation frameworks → goo.gle/4kQREXN
3
7
1
48
5,053
Orr Krupnik retweeted
We’re soon releasing Lightcone, which brings Light to your Mac. Lightcone will be able to automate any task on your computer. It will be powered by our very own pre-trained foundation model, built from the ground up for computer use. lightcone.sh
3
1
16
2,313
Orr Krupnik retweeted
We’re hiring ML Engineers at Tzafon in SF. Join our elite team (IOI, IMO, Google alumni) working at the intersection of AI research and practical engineering. Focus areas include multi-agent RL, memory architectures, efficient sampling techniques, and Large Action Models. Send ML demos or papers to [email protected] or DM directly.
4
12
1,208
Orr Krupnik retweeted
Want to learn / teach RL? Check out new book draft: Reinforcement Learning - Foundations sites.google.com/view/rlfoun… W/ @shiemannor and @YishayMansour This is a rigorous first course in RL, based on our teaching at TAU CS and Technion ECE.
10
115
4
676
82,345
Orr Krupnik retweeted
We recently had a bit of a breakthrough in generalization in RL, led by @EvZisselman TL;DR: learning MaxEnt exploration generalizes better than maximizing reward. We use this to set a new SOTA for ProcGen + significantly improve on hard games like Heist! #NeurIPS2023 Details👇
3
15
1
132
19,786
We are excited to announce the Technion Robotics Seminars (TRS) resume in Fall'21 semester, see attached. Co-organized with Oren Salzman, Erez Karpas (@karpase), Amir Degani, and Aviv Tamar(@AvivTamar1).
2
1
13
Excited to share our new #ICRA2021 paper on “Contrastive Domain Randomization” (CDR) with my amazing collaborators @nikogrupen, and @AvivTamar1: Website: sites.google.com/view/contra… Paper: arxiv.org/abs/2103.11144 Code: github.com/carmelrabinov/cdr #robotics #ContrastiveLearning #AI
1
4
3
21
Israeli Reinforcement Learning Day #IRLD2021 will take place on April 7th with amazing keynote speakers @AvivTamar1, Yishay Mansour and @MannorShie and great talks! Registration is free, event link will be sent to all *registered* participants on April 6th. See you there!
1
1
7
Orr Krupnik retweeted
Replying to @SoloGen
Awesome! I'm also teaching RL this semester (my first time teaching in a hybrid zoom+live class mode), and we're also following a more mathematical approach, though it's advanced ugrad so a bit slower (avivt.github.io/avivt/teachi…)
1
3
1
13
Orr Krupnik retweeted
Soft-IntroVAE accepted as oral at CVPR! Great work by @TalDaniel8 TL;DR: stable VAE with adversarial component that's great at generation and inference. Use VAEs in your code? Consider switching to S-IntroVAE. Paper + excuses (tweet->blog post->code tutorial) below
Excited to share our new work on (introspective) VAEs! We introduce Soft-IntroVAE. With @AvivTamar1 Paper: arxiv.org/abs/2012.13253 Website: taldatech.github.io/soft-int… Code: github.com/taldatech/soft-in… [1/7]
2
2
11
Orr Krupnik retweeted
Our 2 ICRA submissions and 1 CVPR submission all got in! What a great way to start the week!
3
1
73
Orr Krupnik retweeted
מלמדים ביולוגיה לתיכוניסטים? שמתי לב שהתלמידים שלי יכולים להפריך את רוב הטענות נגד החיסון לקורונה נטו על בסיס הידע הביולוגי שהם צברו במחצית הראשונה של כיתה י'. באופן טבעי, זה גרם לי לבנות מערך שיעור קצר (כחצי שעה) שבו נעלה טענות לגבי החיסון והם יסבירו למה הן לא נכונות. >>
14
37
3
471
We are proud to announce that #Beresheet2 is going to head to the Moon in the first half of 2024! It will include two landers, each of which will carry out experiments on the surface of the Moon, and an orbiter that will stay for several years. #IsraelToTheMoon
47
209
115
1,026
[1/3] Excited to share Subgoal Trees (SGT) in #icml2020! SGT is an alternative decomposition of the multi-goal RL problem based on a divide-and-conquer idea: given start and goal states predict a subgoal and solve two simpler problems (start->subgoal and subgoal->goal). #icml
1
3
3
5
New blog post reflecting on my research so far and trying to look at the broader picture from a community point of view. tesslerc.github.io/posts/drl… The bottom line is - deep RL works but there are still many basic problems to solve.
3
17
2
65