iiinova — where your digital self keeps evolving

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

030

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!

130

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)

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.

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.

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.

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.

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).

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!

🚨 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

14817

+ 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

011

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.

100

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.

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.

100

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 🏆

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.

Badr AlKhamissi @bkhmsi.bsky.social · 06/05/2026

🧠 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. 🧵

1299

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!

040

Badr AlKhamissi @bkhmsi.bsky.social · 27/04/2026

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!

010

Badr AlKhamissi @bkhmsi.bsky.social · 27/04/2026

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 👇

1112

Badr AlKhamissi @bkhmsi.bsky.social · 22/04/2026

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👇

041

Badr AlKhamissi @bkhmsi.bsky.social · 22/04/2026

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/

130

Badr AlKhamissi @bkhmsi.bsky.social · 17/02/2026

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!

040

We also added a statistics page visualizing key aspects of the community: academia vs. industry, research areas, positions, etc. Webpage: egyptians-in-cs.github.io#/en/stats

The website now has much better filtering by subfield, making it easier to explore different areas of Computer Science and discover researchers working on specific topics.

100

One of the most powerful additions: an interactive map showing where Egyptian researchers are around the world 🌍 It highlights the global Egyptian diaspora, and how widely Egyptian researchers are contributing across the world.

100

Badr AlKhamissi @bkhmsi.bsky.social · 07/01/2026

The website now features 262 Egyptian researchers across all of Computer Science — from systems and theory to AI, security, HCI, and more. What started as a short list became a much broader story about visibility and representation.

100

4/ 🤝 Re-Align 2026 is made possible by an interdisciplinary team of co-organizers: @bkhmsi.bsky.social, Brian Cheung, @dotadotadota.bsky.social, @eringrant.me, Stephanie Fu, @kushinm.bsky.social, @sucholutsky.bsky.social, and @siddsuresh97.bsky.social!

050

Badr AlKhamissi @bkhmsi.bsky.social · 07/01/2026

3/ 🌟 Joining us at Re-Align 2026 is a fantastic lineup of invited speakers covering ML, neuroscience, and cognitive science: David Bau, Arturo Deza, @judithfan.bsky.social, @alonaf.bsky.social, @phillipisola.bsky.social, and Danielle Perszyk!

141

Badr AlKhamissi @bkhmsi.bsky.social · 07/01/2026

🎉 Re-Align is back for its 4th edition at ICLR 2026! 📣 We invite submissions on representational alignment, spanning ML, Neuroscience, CogSci, and related fields. 📝 Tracks: Short (≤5p), Long (≤10p), Challenge (blog) ⏰ Deadline: Feb 5, 2026 for papers 🔗 representational-alignment.github.io/2026/

1169

Badr AlKhamissi @bkhmsi.bsky.social · 31/12/2025

This year was hard, personally and globally, from ongoing visa issues that disrupted my life and may prevent me from achieving one of my dreams, to the state of the world itself. Still grateful for what I achieved and for everyone who supported me. Wishing us all a brighter year ahead. ✨

050

Badr AlKhamissi @bkhmsi.bsky.social · 07/12/2025

Looking forward to be speaking at IndabaX Sudan on Building Responsible and Ethical LLMs! 📅 Saturday, December 13th ⏰ 2:00 PM (GMT+2) Register here: docs.google.com/forms/d/e/1F... See you all there! :)

020

Badr AlKhamissi @bkhmsi.bsky.social · 02/11/2025

On my way to #EMNLP2025 🇨🇳 I’ll be presenting our work (Oral) on Nov 5, Special Theme session, Room A106-107 at 14:30. Let’s talk brains 🧠, machines 🤖, and everything in between :D Looking forward to all the amazing discussions!

190

6/ We also wondered: if neuroscientists use functional localizers to map networks in the brain, could we do the same for MiCRo’s experts? The answer: yes! The very same localizers successfully recovered the corresponding expert modules in our models!

111

5/ One result I was particularly excited about is the emergent hierarchy we found across MiCRo layers: 🔺Earlier layers route tokens to Language experts. 🔻Deeper layers shift toward domain-relevant experts. This emergent hierarchy mirrors patterns observed in the human brain 🧠

100

4/ We find that MiCRo matches or outperforms baselines on reasoning tasks (e.g., GSM8K, BBH) and aligns better with human behavior (CogBench), while maintaining interpretability!!

100

2/ 🧩 Recap: MiCRo takes a pretrained language model and post-trains it to develop distinct, brain-inspired modules aligned with four cognitive networks: 🗣️ Language 🔢 Logic / Multiple Demand 🧍‍♂️ Social / Theory of Mind 🌍 World / Default Mode Network

Badr AlKhamissi @bkhmsi.bsky.social · 02/10/2025

🚀 Excited to share a major update to our “Mixture of Cognitive Reasoners” (MiCRo) paper! We ask: What benefits can we unlock by designing language models whose inner structure mirrors the brain’s functional specialization? More below 🧠👇 cognitive-reasoners.epfl.ch

2319

Excited to be part of this cool work led by Melika Honarmand! We show that by selectively targeting VLM units that mirror the brain’s visual word form area, models develop dyslexic-like reading impairments, while leaving other abilities intact!! 🧠🤖 Details in the 🧵👇

010

8/ 🧬 Brain alignment: Neuroscience localizers (e.g., for language, multiple-demand) rediscover the corresponding experts in MiCRo, showing functional alignment with brain networks. However, ToM localizer fail to identify the social expert. Figures for MiCRo-Llama & MiCRo-OLMo.

7/ 🧩 Steering & controllability: Removing or emphasizing specific experts steers model behavior: Ablating logic expert hurts math accuracy; suppressing social reasoning improves math slightly—showcasing fine-grained control.

6/ 🔄 Interpretable routing: Early layers route most tokens to the language expert; deeper layers route to domain-relevant experts (e.g., logic expert for math), matching task semantics.

5/ 📈 Performance gains: We evaluate on 6 reasoning benchmarks (MATH, GSM8K, MMLU, BBH…), MiCRo outperforms both dense and “general‑expert” baselines: modular models with random specialist assignment in Stage 1.

4/ 📚 Training curriculum (3 stages): • Stage 1: Expert training on small curated domain-specific datasets (~3k samples) • Stage 2: Router training, experts frozen • Stage 3: End-to-end finetuning on large instruction corpus (939k samples) This seeds specialization effectively.

2/ 🔍 Motivation: Humans rely on specialized brain networks—e.g., language, multiple-demand, ToM, default mode—for different cognitive tasks. MiCRo mimics this by dividing transformer layers into four experts. Figure from @evfedorenko.bsky.social's review paper: www.nature.com/articles/s41...