Assistant Professor @imperialcollege MTS @amilabs
London
Joined March 2016
- Tweets774
- Following1.5K
- Followers4.2K
- Likes14.5K
Pinned Tweet
A triple life update:
1. I recently moved from NYC to London
2. I’m starting a new lab at @imperialcollege
3. I’ve joined my friends and mentors at @amilabs as a founding Member of Technical Staff
It's time to build 🏗️🧑🍳
Amir Bar retweeted
behold the anti-schmidhuber. upon contact with schmidhuber, they both undergo instant annihilation and convert their entire mass into pure energy.
We got scooped by this great paper, and I’m not even mad: they did a wonderful job and chose a much better title than I had in mind.
We have StopGrads everywhere and casually gloss over them as if we’re still doing gradient descent. Meanwhile, our parameters follow a vector field that may not even be the gradient of any loss function.
For flow-map learning, they explicitly solve the idealized training dynamics and show convergence to the right solution, despite the absence of an underlying loss.
Really great work by @maxwshen, @marikgoldstein and the team!
arxiv.org/abs/2609.16222
cool. it will be interesting to see in larger scale. Intuitively, we know that random search scales exponentially with the number of parameters whereas sgd is polynomial (for convex objectives).
Join Jesse's tutorial if you're attending NeurIPS!
🎉 Excited to bring our book The Principles of Diffusion Models to @NeurIPSConf 2026 as a tutorial—this time, all about diffusion LLMs! 🚀
🔹 How do ideas from continuous diffusion extend to language generation?
🔹 What are the latest advances and open frontiers in diffusion LLMs?
From foundational principles to cutting-edge research. Stay tuned!
📆 Mon, Dec 7, 9 AM AEDT
📍 TikTok Ent Centre (Sydney, Australia)
w/ organizers @ssahoo_ @mittu1204 @StefanoErmon
NAVA-WAM: Can we learn latent actions for World Action Models?
We revisit a classic idea: jointly learning inverse and forward dynamics models while avoiding the identity shortcut (e.g., @SeonghyeonYe, @garridoq_, @jeffacce, @pulkitology, and others).
The twist: Both models are implemented within a single WAM architecture, using attention masks!
To prevent the identity shortcut, the latent actions can only sees a noisy version of the next frame, forcing the latent action to capture useful dynamics rather than simply copying the future.
This simple pretraining strategy improves both performance and efficiency across robotics benchmarks!
🧵👇
🤖 Abundant videos capture how the world works, but robot policies still rely on costly action labels to learn.
🔥 Excited to share our work at @Meta:
📢 Native Action-Prior Learning from Videos for World Action Models 📢
📄arxiv.org/pdf/2610.03391
🌐zhaochongan.github.io/NAVA-W…
🧵
𝗣𝗿𝗼𝗪𝗔𝗠 imagines latent visual subgoals to guide action prediction. Strong performance across robotics benchmarks and real-world experiments.
🧵👇
👋 𝗛𝗲𝗹𝗹𝗼 𝗧𝘄𝗶𝘁𝘁𝗲𝗿! 𝗙𝗶𝗿𝘀𝘁 𝘁𝘄𝗲𝗲𝘁 𝗮𝗹𝗲𝗿𝘁 🚨
Why not give it to a paper!
Excited to share our work at @Meta: 𝗣𝗿𝗼𝗪𝗔𝗠
In this paper, we try to answer:
"How should a robot imagine the future before it acts?"
🌐 Webpage: sii-ferenas.github.io/ProWAM…
🧵👇
FAR is a new memory framework for world models. Retrieval from memory utilizes different sensory cues (vision, audio, etc). Draws inspiration from recent ideas in LLMs (e.g RAG).
See 🧵👇
(1/9)🧠World models don’t just need more memory. They need the right memory to predict what happens next.
We introduce FAR (Future-Aware Recall), a principled framework that learns episodic recall for world models:
🏆Beats hand-designed recall—even with the same cues
🎛️Automatically learns which recall cues to trust for each query, among all available cues
🔄Recalls the right state, even as the environment changes
Huge thanks to my collaborators @JCJesseLai, @nguyencongbacbk, @_amirbar, Jong Chul Ye, and @mittu1204 for making this work possible!
Amir Bar retweeted
(1/9)🧠World models don’t just need more memory. They need the right memory to predict what happens next.
We introduce FAR (Future-Aware Recall), a principled framework that learns episodic recall for world models:
🏆Beats hand-designed recall—even with the same cues
🎛️Automatically learns which recall cues to trust for each query, among all available cues
🔄Recalls the right state, even as the environment changes
Huge thanks to my collaborators @JCJesseLai, @nguyencongbacbk, @_amirbar, Jong Chul Ye, and @mittu1204 for making this work possible!
Another option is to end double-blind review. It used to work well, but it removes accountability. Authors who submit AI slop should take the reputational hit, very simple.
More complex rules just lead to unintended consequences and are just too difficult to enforce.
Exactly, tokens are now cheap and like Yi reminds us, intelligent agent should be able to compress.
Concretely:
1. Desk rejection phase based on abstract only
2. Shorter papers
3. Cancel author-reviewer discussion phase, it’s now all AI slop anyway.
Amir Bar retweeted
Given what we know about AIs breaking human rules/ethics to achieve a goal, this can only end badly. When the success metric is about getting published and there is no reputational disadvantage to cheating, cheating will happen. When AIs get to review their own papers, guess what they will do.
Amir Bar retweeted
🚨 Call for participation: the Child–Robot Dexterity Challenge at the CoRL 2026 𝗦𝗣𝗜𝗡 𝗪𝗼𝗿𝗸𝘀𝗵𝗼𝗽! Bring your robot, or run on ours. Apply by Oct 5.
🧒🤖 Robots and a child will take on the same everyday tasks, side by side, live on stage. The demonstration offers a simple way to see what today’s robots can already do and where there’s still room to improve.
We make it easy to join:
- Bimanual YAM setup (MolmoAct 2 style) + a dexterous Sharpa hand setup, provided on stage.
- Bring your own hardware (subject to approval).
🌐 Website: spin-workshop.github.io/#con…
🗣️ Speakers & panel from Stanford, Columbia, TU Darmstadt, and AMI Labs.
📍 Nov 12 · Austin, TX
#CoRL2026
Amir Bar retweeted
12k RoboCap units sold to-date.
There'll be millions of hours of ego data collected on RoboCap by our customers/partners and @BitRobotNetwork
1/ Better robot models come from better data.
That takes two things: hardware that captures it fully, and tooling that turns it into training signal.
We built the first. Today, @GroundedSI launches the second: Grounded API. Access both hardware + API ↓
Isn't this obvious though? CoT from a previous model can provide supervision for the next generation model.
Amir Bar retweeted
🌀 One day to go!
Join WOOP tomorrow from 13:30–18:00 at #ECCV2026.
🎙 Invited talks: @RanjayKrishna, @vincesitzmann, @lealtaixe, @CarlDoersch, @_amirbar
🗣 Panel discussion: @vincesitzmann, @CarlDoersch, @_amirbar, and @taiyasaki
🔗 eccv26woop.github.io/
WOOP @ #ECCV2026 !
As part of the ECCV workshop on World Models in the Loop, we are inviting submissions of papers to our Non-Proceedings track, accessible via eccv26woop.github.io/#call
Organizers: @NhiPham49 @KenJSimmoteit @thunguyenphuoc @AdamKortylewski
See you in Malmö!
Amir Bar retweeted
Happy to co-host the 2nd edition of the FOUND Workshop in @eccvconf.
For this edition, we are heavily emphasizing on Physical AI and World Models, with a specific focus on data curation for real-world embodiment.
Join us on 9th Sept 8:30 to 12:30 at MalmöMässan E1. More info⬇️
Amir Bar retweeted
We are extending the submission deadline of our Continual World Models workshop by one week.
If you’ve been considering submitting, we hope the extra time gives you a chance to share your work with us.
A world model should change when the world does.
I’m helping organize Continual World Models at #NeurIPS2026 in Sydney this December. We’re looking for models that keep learning after deployment.
Submissions close Aug 29, AoE.
continual-world-models-works…
