Mandana Samiei

PhD candidate @Mila_Quebec @rllabmcgill studying reinforcement learning and the brain. Board of Directors @WiMLworkshop. Previously @ThinkSurgical

[Mandana Samiei @MandanaSamiei]

While the top humans still outperform models in proposing interventions that bear the greatest cumulative information gain and most efficient elimination of hypotheses, most models approximate average human performance.

[Abbas Mehrabian @abbas_mehrabian]

Yesterday OpenAI published hundreds of new math results. Among them was a groundbreaking algorithm to multiply matrices, showing ω⩽2.25. Our team had the best bound before OpenAI's result. Below I have explained the history of ω and why we care about it: https://t.co/f8vxstCuuE

[Sinead Williamson @ COLM 2026 @sineadwilliamso]

For a while now, I’ve been curious about LLMs' verbalized probabilities. What are they actually tracking? In our new paper https://t.co/PKSAh1cWXm , we’ve uncovered some surprising findings🧵

[Siva Reddy @sivareddyg]

Excited to share that I have joined @MistralAI, where I will be building a team in Montreal and across Canada to push the frontier of safe, capable, and open LLMs.

Mistral has played, and will continue to play, an important role in keeping frontier AI open and advancing it safely. I deeply believe in Mistral's mission, and since joining, that conviction has only grown after seeing firsthand the ambition and commitment of the people here.

A big part of my role will also be building strong bridges between Mistral and @McGillU, @Mila_Quebec, and the broader Montreal and Canadian AI ecosystem.

And we are hiring! Across levels and across the LLM stack. If you are excited about pushing the frontier, reach out.

[Mandana Samiei @MandanaSamiei]

I want to echo Sara’s words! @hugo_larochelle is an amazing researcher and such a thoughtful person. His lecture on meta-learning was a big inspiration for me.
We’re grateful to have him leading Mila’s scientific direction and helping shape the community we’re all part of!

[Sara Hooker @sarahookr]

Very happy to see this profile of @hugo_larochelle

Before I join Google Brain, I watched all his lectures from Sheffield. You have to remember at the time, all of the knowledge for deep learning belong to less then 100 people. He made it available to many more.

He became my PhD advisor many years later. Hugo is an exceptional researcher, has shaped so much of our ecosystem.

[Tatjana Chavdarova @T_Chavdarova]

Realized I never shared this here 😊

Earlier this year, I chatted with @TU_Wien Informatics about my research, my path into computer science, and what I enjoy most about it.

Very happy to be part of the TU Wien community!

https://t.co/706TY9wqC3

#AI #MachineLearning

[Théo Vincent @Theo_Vincent_]

🏅Gradient Iterated Temporal-Difference Learning received Best Paper Award for empirical RL research @RL_Conference🥳

[Mandana Samiei @MandanaSamiei]

Really excited to see DiG-bench out!

Discovery is such a fundamental capability for intelligent agents: not just answering questions, but figuring out what to test, what evidence matters, and how to update from experience.

[James Whittington @jcrwhittington]

We’re excited to announce DiG-bench, a new benchmark for discovery!

Over the last few weeks we’ve been testing frontier AI models on our novel discovery games and seeing how they score. Each game is a text-based environment, so they probe discovery capabilities in the natural domain of language models, rather than requiring additional, potentially confounding, visual understanding.

TL;DR frontier models have improved a lot over the last few months. But they are still stumped by some surprisingly simple problems, even in their native text domain.

With

@cocosci_lab (@Princeton) @MITCoCoSci (@MIT) @SchmidhuberAI (@KAUST_News) @misovalko (@Inria) @tri_dao (@PrincetonCS) @RMBattleday @zebkDotCom @FraserGreenlee @akaijsa @ClareMaguire @TimMuller1 @kubicek_ales @physicscat0x7d @SukritSumant @thoughtchannel_

(1/5)

[Grigory Sapunov @che_shr_cat]

1/Again about continual learning.

What if catastrophic forgetting isn't a bug, but a mathematical necessity?

We build complex systems to retain old knowledge, but new theory proves that in non-stationary streams, perfect memory is an optimization liability. Here is why. 🧵

[Mandana Samiei @MandanaSamiei]

Presenting our work today at #CogSci2026 🎉

Our findings reveal interesting similarities, and important differences, in how humans and modern AI systems explore causal structures.

If you're at the conference, come by and say hello!

Today 5:30–7:00pm Reasoning session, P3-L-104

[Mandana Samiei @MandanaSamiei]

We recently investigated how humans and AI reason through causal rules: https://t.co/1NigtmCyKS

We gave adults and LLMs agency to actively experiment using a Blicket causal game, breaking down their process into the iterative loop shown in the following figure:

[Alison Gopnik @AlisonGopnik]

My MIT commencement speech now online. I worked very hard on this, trying to put together the MIT scientists, children, human evolution,caregiving grandmoms and the Lunar Society - I hope it provides some hope and inspiration in a hard time for science.

https://t.co/mGyUtLplwv

[WiML @WiMLworkshop]

📣 Applications are now open for the NeurIPS 2026 Affinity Support Pilot Program! ✈️🏨🎟️🛂 Support available

📅 Deadline: July 20, 2026 EOD (AoE)
🔗 https://t.co/tCRUe6RyV2

#WiML #NeurIPS2026

[ICML Conference @icmlconf]

🏆Finally, the Test of Time Award:

Asynchronous Methods for Deep Reinforcement Learning

Originally published at #ICML2016

[Darshan Patil @dapatil211]

Final takeaways: Adaptation is not only about the final objective. The path into that objective matters.

Abrupt adaptation can distort useful learned features. Loss smoothing gives the model a softer transition from source to target training, improving the retention/adaptation tradeoff while still allowing specialization.

Check out our paper for more results!

Thanks to my amazing collaborators @KateLobacheva, Razvan Pascanu, and @apsarathchandar.

7/7

[Jesse Farebrother @JesseFarebro]

Come hang out with @ahtouati & I at our #ICML2026 poster on Wednesday afternoon to learn about jumpy world models!

[Ahmed T. @ahtouati]

How do we get RL agents to plan over long horizons without drowning in compounding errors?

Instead of reasoning over primitive, step-by-step actions, what if they could compose pre-trained policies across multiple timescales?

Poster #1117 Wed 8th at 2:30 pm #ICML2026

[WiML @WiMLworkshop]

Excited to announce the WiML Symposium @ ICML 2026 Evening Social! 🎉

📅 Tuesday, July 7
⏰ 7:00–10:00 PM KST
📍 Gangnam, Seoul
👉 Register here: https://t.co/D9tHjHsGG9

Looking forward to seeing many familiar and new faces there. See you in Seoul!

#WiML #ICML2026

[Anthony GX-Chen @AntChen_]

A great collaboration co-led by @MandanaSamiei and @eunice_yiu_

What I found interesting:
- Per-action info gain analysis (from our prev work 👇) applied to human adult data
- Human adults are better than prev thought esp if they can actively explore!

https://t.co/BHyDoNtINm

[Anthony GX-Chen @AntChen_]

Check out our new work on how humans & LLMs causally reason and explore 🧐, with freshly collected human adult data!

Some takeaways:
- Both show a bias for disjunctive rules
- LLMs still lag behind the best humans in info gain per action

Accepted at #CogSci2026 🇧🇷

[Dongyan Lin @dongyanl1n]

(1/n) Excited to share our new paper from @AIatMeta FAIR: "EgoBabyVLM: Benchmarking Cross-Modal Learning from Naturalistic Egocentric Video Data" 👶Human infants learn language from sparse, noisy multimodal input. Today's VLMs can't. We built a benchmark + challenge to close that gap.
w/ @rust_phillip, @angelvillar96, @juanmiguelpino, @mcxfrank, Emmanuel Dupoux and many other amazing collaborators. 🧵

[Mandana Samiei @MandanaSamiei]

If you're excited about understanding how frontier models adapt, generalize, and acquire capabilities, I can't think of a better person to work with. Go join his lab!

[Marius Mosbach @ COLM 2026 🌉 @mariusmosbach]

🚨 I’m happy to share that I'm joining Saarland University as a tenure-track Professor in the Department of Language Science and Technology, and the German Research Center for AI (@DFKI) as a Scientific Director, starting October 2026! 🚨 1/6