Badr AlKhamissi

@gretatuckute.bsky.social

Egyptians in CS Research now features 300+ researchers. Hoping it inspires students from underrepresented backgrounds by showing them what a path into research can look like. If you know someone missing from the list, nominate them. egyptians-in-cs.github.io

Badr AlKhamissi reposted

Greta Tuckute

17 Sept

Thanks so much for the fun conversation @wiair.bsky.social ! A pleasure to chat about language, LLMs, and memory--covering some work with @bkhmsi.bsky.social @mschrimpf.bsky.social @michael-lepori.bsky.social @klemenkotar.bsky.social @evfedorenko.bsky.social @thomashikaru.bsky.social, among others!

Women in AI Research - WiAIR@wiair.bsky.social

New episode of #WiAIRpodcast: Greta Tuckute @gretatuckute.bsky.social talks to @j-novikova-nlp.bsky.social about what brain-LLM alignment does and does not tell us, and why, asked to keep one label out of neuroscience, cognitive science and AI, she drops AI first

8 Sept

So do LLMs really think in English? 🤔 Turns out it depends on how you ask. Different latent-language-ID methods give you different answers! We test 27 candidate languages across 12 diverse models, two of them analyzed through pretraining. Led by the amazing @bayazitdeniz.bsky.social 👇

Deniz Bayazit@bayazitdeniz.bsky.social

1/🚨 New paper #emnlp2026 Do multilingual #LLMs work in English? It depends on how you ask. In "Lingua Franca or Probing Artifact?", we probe the same hidden states with three latent language identification methods and get different answers. 🧵👇 #interpretability #NLProc

arxiv.orgLingua Franca or Probing Artifact? Rethinking Latent Language in Multilingual LLMsLatent language identification is often used to argue that multilingual language models route computation through language-specific states, such as English pivots. However, existing probes infer laten...

8 Sept

One direction I'm especially excited about: relating latent language to the brain's "language of thought," which recent work by @hopekean.bsky.social, @evfedorenko.bsky.social et al. suggests isn't natural language at all. 🤯 Kean et al.: www.pnas.org/doi/10.1073/... Ours: arxiv.org/abs/2609.00155

Badr AlKhamissi reposted

Hannes Mehrer @CCN2026

@hannesmehrer.bsky.social

4 Aug

"So, you have a highly predictive model, now what?" At this #CCN2026 event on Wed, 5 Aug, 4pm at NYU Skirball Theatre I will present work on using ANN brain models for visual prostheses (tinyurl.com/modelguidedm...) and for dyslexia (tinyurl.com/neuroaidysle..., first author: Melika Honarmand).

3 Aug

Excited to be at #CCN2026 in NYC! I'll be presenting our spotlight poster, Topo-Omni, with @hannesmehrer.bsky.social 🧠 📍 Poster F14 — Session F 🗓️ Thursday, Aug 6 Come by if you're around, we'd love to talk about how our multimodal model can discover functionally selective brain regions!

3 Aug

You can find more about our paper here: Link: topo-omni.epfl.ch Summary: bsky.app/profile/bkhm...

Badr AlKhamissi@bkhmsi.bsky.social

🚨 New Preprint! 🧠 We gave an AI model one simple rule: rearrange your neurons so that nearby ones respond alike. We never told it what a face, a voice, or a sentence was. It grew brain-like maps for all three anyway. 🧵👇 🌐 Website: topo-omni.epfl.ch

bkhmsi.substack.comAI and Neuroscience: Still a Virtuous Circle?Neuroscience once handed AI its best ideas. Today the current runs mostly the other way.

30 July

Finally wrote up something I've been thinking about for a while, and with CCN next week, feels like the moment to share it: is the "virtuous circle" between AI and neuroscience that Hassabis et al. described in 2017 still turning? Curious what you think! bkhmsi.substack.com/p/ai-and-neu...

13 July

Atari from raw pixels. AlphaGo vs Lee Sedol. AlphaFold. DeepMind has inspired me for years, and its neuroscience work is why I got into NeuroAI. So it means a lot to say I've joined Google DeepMind in Zurich as a Student Researcher for the next few months. In Zurich? Say hi :)

Badr AlKhamissi reposted

Yingtian (David) Tang

@davidtyt.bsky.social

7 July

🚨 NEW PREPRINT Videos strongly shape activity across the visual cortex. But can we design videos that maximally drive specific brain regions? We present NEvo 🧬🧠 — a neural-guided evolutionary framework that synthesizes videos to maximally activate target visual ROIs. (1/10)

Badr AlKhamissi reposted

Abdulkadir Gokce

@akgokce.bsky.social

2 July

🧠 What's the best way to spend compute & neural data to build brain-aligned models? Our new ICML 2026 paper charts the scaling laws across 8 neural datasets & 600+ vision models. 👇 🌐 multimodal-brain-scaling.epfl.ch w/ @davidtyt.bsky.social & @mschrimpf.bsky.social

actu.epfl.chEPFL researchers create an AI model that thinks like we doAn EPFL team has created a new Large Language Model that is structured similarly to a human brain, allowing users more control and moving away from “black box” AI.

26 June

Our work on Mixture of Cognitive Reasoners (MiCRo) made EPFL's front page today! 🎉 Read the full story here: actu.epfl.ch/news/epfl-re... Grateful to everyone involved!

Badr AlKhamissi reposted

Yingtian (David) Tang

@davidtyt.bsky.social

26 June

🧠🎬 RELEASE We are releasing two Hugging Face model cards for video brain encoding. 🎥 short video clips → 🧠 brain responses MODEL 1: vjepa2-basic 🔗 huggingface.co/epfl-neuroai... MODEL 2: vjepa2-enhanced 🔗 huggingface.co/epfl-neuroai... Brain encoders. Off the shelf !

huggingface.coepfl-neuroai/vjepa2-encoder-basic · Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science.

Badr AlKhamissi reposted

Martin Schrimpf

@mschrimpf.bsky.social

cell.comA unifying framework for functional organization in early and higher ventral visual cortexMargalit et al. develop a topographic artificial neural network that predicts both functional responses and spatial organization of multiple cortical areas of the primate visual system. In turn, the m...

A wiring-cost-like loss has been shown to induce brain-like topography in vision (eg www.cell.com/neuron/fullt... and language (e.g. topolm.epfl.ch, toponets.github.io). The same spatial loss yields cognitive clusters in a multimodal model which even discovers new ones! topo-omni.epfl.ch #NeuroAI🤖🧠🧪

Badr AlKhamissi@bkhmsi.bsky.social

10/ One spatial-smoothness rule, applied to a pretrained multimodal model, recovers cortex-like organization + generates new, testable hypotheses about the brain. co-led w/ @hannesmehrer.bsky.social, collab w/ Lara Marinov, Ahmed Abdelaal, @akgokce.bsky.social & @mschrimpf.bsky.social!

11/ Links 📄 Paper: arxiv.org/abs/2606.09770 💻 Code: github.com/epflneuroailab/topo-omni 🤗 Model: huggingface.co/epfl-neuroai/topo-omni 🌐 Website: topo-omni.epfl.ch

8/ The big one: discovering NEW brain networks We used Topo-Omni's sheet to screen video clips and found two clusters with no clear precedent in the literature: 🦅 An animals network (snakes, birds, big cats) 🏔️ A natural landscapes network (beaches, mountains, terrain)

9/ We tested these AI-derived predictions on independent human fMRI (Spacetop) and found right-lateralized prefrontal activation for both. An AI model just predicted new organization in the brain. 🤯

6/ "But doesn't this hurt performance?" We checked. Vs. non-topographic baselines (same backbone, same task loss): ✅ Brain predictivity statistically tied on 11/12 visual ROIs (Natural Scenes Dataset) ✅ Best overall OmniBench accuracy Topography is basically free.

7/ Causal Tests We can also drive or suppress clusters: 🔼 Drive face units → model "sees" faces almost everywhere 🔽 Suppress top 10% of face units → face recognition collapses to ~0%, while bodies/scenes/objects stay intact These regions aren't just correlated, they're causally responsible.

4/ Audition In the audio encoder: a speech-selective region tracks the human superior temporal gyrus (r=0.69), it responds to intelligible speech but drops for "quilted" speech. A separate voice-selective region prefers human vocalizations, like the temporal voice areas.

5/ Cognition Feed it text, and the deeper module finds THREE networks using the localizers we tested: 🗣️ Language (sentences > nonwords) 🧮 Multiple-demand (math > sentences) 🤔 Theory of mind (false-belief > false-photo) One spatial rule → dissociable cognitive networks. No brain data, no labels.

2/ Like the real cortex: one continuous sheet, from retinotopy to the language network. Most topographic models fragment this, a separate 2D map per layer, and for one modality only. Topo-Omni instead puts vision, audio, AND cognitive units of Qwen2.5-Omni on ONE shared sheet!

3/ Vision Result: in the vision encoder, distinct patches emerge for faces, scenes, objects, and visual word forms: echoing the FFA, PPA, LOC and VWFA of your visual ventral stream. Their response profiles correlate with real human fMRI (r up to 0.89).

mirage-brain.epfl.chMIRAGE: Adaptive Multimodal Gating for Whole-Brain fMRI Encoding

4 June

So why build toward a brain-encoding foundation model? → Simulate fMRI responses to sensory stimuli → Insights about the brain → Path toward clinical applications MIRAGE is our first step. 🧠 More on the project site: mirage-brain.epfl.ch w/ @akgokce.bsky.social & @mschrimpf.bsky.social

Badr AlKhamissi@bkhmsi.bsky.social

🧠 When you watch a movie, your brain blends sight, sound, and speech into a single experience. Should models of the brain blend them too, or keep the senses separate until the very end? We built MIRAGE to find out. It sets a new SOTA for predicting whole-brain fMRI from movies. 🧵

Badr AlKhamissi reposted

Martin Schrimpf

@mschrimpf.bsky.social

2 June

Two simple ideas for building improved brain encoding models: 1. learn to use representations from all model layers via a gating mechanism + 2. start from natively multimodal features for multimodal predictions. State of the art performance; see mirage-brain.epfl.ch for details #NeuroAI 🧠🤖🧪

Badr AlKhamissi@bkhmsi.bsky.social

Each modality also traces a distinct anatomical pattern: vision → occipitotemporal, audio → auditory cortex, text → the language network. MIRAGE's largest gains over our linear baseline land in visual & dorsal-attention areas, exactly where rich social-movie content demands integration.

1 June

+ we've a demo: play a clip and watch MIRAGE's predicted whole-brain activity light up in sync 🧠🍿 📄 Paper: arxiv.org/abs/2605.29850 💻 Code: github.com/epflneuroail... 🤗 Model: huggingface.co/epfl-neuroai... 🌐 Demo: mirage-brain.epfl.ch Joint work w/ @akgokce.bsky.social & @mschrimpf.bsky.social

And this isn't a quirk of one model. Across 2 backbone families and 3 scales, native fusion wins at every single scale. Fusing modalities during pretraining yields features that are more brain-aligned than stitching unimodal streams together afterward.

1 June

Bonus: MIRAGE is inspectable. The gates' attention weights reveal which backbone layers each modality reads from. 👀 Vision is sharply tuned to mid-depth layers (~25–30), text spreads across mid-to-late layers, audio is the most diffuse.

Enter MIRAGE 🪄 Most encoding models pin a linear readout to one fixed layer. MIRAGE does neither. A frozen omni-modal backbone (Qwen3-Omni) exposes all 48 layers → per-modality cross-attention gates pool them adaptively → a transformer maps to cortex non-linearly, with a per-subject head.

1 June

Key finding: native fusion beats post-hoc fusion at every architectural level: linear ridge, brain encoder, and full MIRAGE. The kicker: on out-of-distribution movies, a single MIRAGE model beats TRIBE v1's 1,000-model ensemble!! Giving a new SOTA on Algonauts 2025 OOD 🏆

Most brain-encoding pipelines pull vision, audio, and language features from separate models, then fuse them late, at the readout. But modern foundation models fuse modalities during pretraining. Which kind of fusion is actually more brain-relevant?

6 May

Crossed 1,000 citations on Google Scholar today, with an h-index of 15 🎈 What felt like a distant goal a few years ago is now real, and I'm especially proud that the bulk of it is carried by my first-author papers. On to the next milestone!

27 Apr

Good morning #ICLR2026 ☀️ The Re-Align workshop kicks off in under an hour! A full day on what we can actually do with representational alignment, from brains to language models to agents. Schedule in the thread 👇

27 Apr

Today's lineup 📅 Invited talks from David Bau, Valentina Pyatkin, Arturo Deza, Alona Fyshe, and Danielle Perszyk, plus contributed talks, posters, and a panel. See you there!

22 Apr

Excited to be in Rio for #ICLR2026 🇧🇷 I'll be presenting our work, Mixture of Cognitive Reasoners (aka MiCRo), on Friday at Pavilion 3, 10:30 AM (#1610). Come say hi :D Happy to chat about NeuroAI, representational & cultural alignment, and/or test-time learning 🧠

22 Apr

I’m also co-organizing the #ICLR2026 Re-Align workshop on April 27th. We've got an amazing lineup of speakers spanning CogSci, Neuroscience, and ML! 🧠🤝🤖 Link: representational-alignment.github.io/2026/

@peterhenderson.bsky.social

22 Apr

Huge shoutout to my amazing co-authors on Mixture of Cognitive Reasoners: @cndesabbata.bsky.social, @gretatuckute.bsky.social, @eric-zemingchen.bsky.social, @mschrimpf.bsky.social, and @abosselut.bsky.social! Link: cognitive-reasoners.epfl.ch Poster below👇

Badr AlKhamissi reposted

Peter Henderson

11 Mar

The current direction of AI labs is “we’re building something that’s going to replace you and we have no plan to make sure you’re going to land in a better place, but we’ll make billions.” The logical reaction is, “shut it down.” Labs need to get serious on addressing labor impacts.

17 Feb

Had the pleasure of being an invited speaker at the KAUST Rising Stars in AI Symposium last week. Inspiring to meet such a talented group of fellow rising stars and learn about their exciting work. Thanks to the KAUST team for a wonderful event!

huggingface.coRe-Align Hackathon Leaderboard - a Hugging Face Space by representational-alignmentSubmit Blue/Red hackathon JSON and rank by alignment scores.

🚀 The Re-Align Challenge is now LIVE! We’re inviting you to explore what properties of vision models and data lead to convergences and divergences in representational alignment. 🔗 Get started: huggingface.co/spaces/repre... 🧵👇

5/ 🏆 In addition to leaderboard winners, we’ll present Outstanding Challenge Report Awards for submissions that are innovative, unusual, or make novel contributions to representational alignment research. Accepted reports will be compiled into a blog post by the organizers.

6/ We’re excited to see creative, rigorous, and unexpected ideas. Let the challenge begin 🔥

huggingface.coRe-Align Hackathon Leaderboard - a Hugging Face Space by representational-alignmentSubmit Blue/Red hackathon JSON and rank by alignment scores.

3/ 📏 Evaluation uses Centered Kernel Alignment (CKA) on embeddings from a fixed held-out image set. You’ll choose from a pre-curated set of models and datasets — but you’re free to use any dataset or metric to guide your search.

4/ 📝 Complete submission = 2 steps: 1️⃣ Submit your list (max 20 models or 1,000 images) to the leaderboard. 2️⃣ Submit a Markdown-formatted Challenge Report on OpenReview describing your approach, analyses, and insights. ⏰ Report deadline: February 26

huggingface.coRe-Align Hackathon Leaderboard - a Hugging Face Space by representational-alignmentSubmit Blue/Red hackathon JSON and rank by alignment scores.

2/ 🎯 Two submission tracks: 🔵 Blue Team — Find a set of maximum 20 models with the most aligned representations. 🔴 Red Team — Find a set of maximum 1,000 images that lead to maximally divergent model representations.