iiinova — where your digital self keeps evolving

Reposted by Victor Morand

MLIA ISIR @mlia-isir.bsky.social · 29/09/2026

Congratulation Mathias Vast on successfully defending your PhD, entitled "Improving the Robustness, Efficiency and Effectiveness of Neural Rankers in Information Retrieval" ! 🎉 Farewell, best of luck in your new journey :)

061 6

Victor Morand @victormorand.eurosky.social · 01/09/2026

6/6 📄 Read the full paper and check out the code to try it in your IR pipeline: - 💻 Code: github.com/xpmir/mice - 📝 Paper: arxiv.org/abs/2602.16299

github.com

GitHub - xpmir/mice: MICE - Minimal Interaction Cross Encoders

MICE - Minimal Interaction Cross Encoders. Contribute to xpmir/mice development by creating an account on GitHub.

100 1

Victor Morand @victormorand.eurosky.social · 01/09/2026

5/6 ⚖️ New Pareto Frontiers Even without pre-computing doc states, MICE consistently pushes the Pareto efficiency–effectiveness frontier beyond traditional cross-encoders. Imagine with pre-computed docs... we are bridging the gap with late-interaction rankers ! (more soon 👀)

120 2

Victor Morand @victormorand.eurosky.social · 01/09/2026

4/6 ⚡ Massive Efficiency Gains - in short: Runs at ColBERT-like latency with cross-encoder expressiveness - Up to 2.5x theoretical FLOP reduction (already ~2x real speedup with our implementation) - Up to 4x latency speedup when pre-computing document representations offline

100 1

Victor Morand @victormorand.eurosky.social · 01/09/2026

3/6 🐭 Enter MICE We turned these insights into a new architecture 1. Mid-Fusion: Q and D encoded separately in early layers 2. Light Cross-Attention between frozen doc states towards query tokens; (drops costly doc MLP/SDPA) 3. Layer Pruning: Drop superfluous top layers

100 1

Victor Morand @victormorand.eurosky.social · 01/09/2026

2/6 🔍 Causal Masking We mask attention blocks inside standard CE, to see what interactions are superfluous. Notably: - Docs don’t need to attend to the query - Q/D can be contextualized independantly until ~ half of the model depth This confirms previous interp works on CE

100 1

Victor Morand @victormorand.eurosky.social · 01/09/2026

🐭 Excited to share MICE (Minimal Interaction Cross-Encoders) - accepted to #EMNLP2026 main ! Cross-encoders are sota for ranking, but often too heavy We ask a simple question: How many operations can be pruned while maintaining effectiveness? 👉 A lot more than you think 🧵

121 2

Victor Morand @victormorand.eurosky.social · 06/07/2026

Presenting ToMMeR at #ACL2026 in San Diego today ! If you are interested in extracting the capacity of LLMs to recognize mentions of any Entity / Concepts with a lightweight probing model, feel free to come in the Grand Hall to see the poster ☺️ bsky.app/profile/vict...

010 1

Reposted by Victor Morand

MLIA ISIR @mlia-isir.bsky.social · 24/04/2026

🎉 Accepted paper at #ICLR2026 🥳 We are proud to announce that Willy Chung, from our team, is an author of: VL-JEPA: Joint Embedding Predictive Architecture for Vision-language by D. Chen, M. Shukor, T. Moutakanni, W. Chung, L. Yu, T. Kasarla, Y. Bang, A. Bolourchi, Y. LeCun and P. Fung. ⬇️Details

121 2

Victor Morand @victormorand.eurosky.social · 22/04/2026

The full story 👨‍💻 Code: github.com/VictorMorand... 📄 Paper: arxiv.org/abs/2510.19410 🤗 We publish all trained models on @huggingface.bsky.social : huggingface.co/models?other... Joint work with Nadi Tomeh, Josiane Mothe, and Benjamin Piwowarski. #NLP #Interpretability #LLM

github.com

GitHub - VictorMorand/llm2ner: Codebase for ToMMeR – Efficient Entity Mention Detection from Large Language Models

Codebase for ToMMeR – Efficient Entity Mention Detection from Large Language Models - VictorMorand/llm2ner

000 0

Victor Morand @victormorand.eurosky.social · 22/04/2026

How can this be useful ? - ToMMeR offers a lightweight, plug-and-play first-stage filter for entity-aware chunking/retrieval, downstream typing... - It can even be used in a **streaming extraction**, identifying entities alongside LLM generation with almost zero latency penalty,

100 1

Victor Morand @victormorand.eurosky.social · 22/04/2026

Theoretical Insight the study provides concrete evidence that mention detection emerge naturally from the language modeling. We found that diverse models (14M to 15B params) converge on a shared notion of entity boundaries (DICE >75%), regardless of scale or architecture

100 1

Victor Morand @victormorand.eurosky.social · 22/04/2026

How well does ToMMeR performs ? 📈**93% recall** zero-shot across 13 diverse NER benchmarks 🎯 **90% Precision**: Validated via a human-calibrated LLM-judge protocol - we aren't just over-predicting. 🚀 **+42x Faster** than generative prompting: ToMMeR extracts mentions in a single parallel pass.

100 1

Victor Morand @victormorand.eurosky.social · 22/04/2026

➡️ We introduce ToMMeR, a probe no heavier than a single attention head (<300K parameters). You can plug in ANY frozen LLM. It extracts these latent signals in a single parallel pass—no schema nor generative prompting required. 🧵

100 1

Victor Morand @victormorand.eurosky.social · 22/04/2026

The Problem: Mention detection —identifying text spans that refer to objects or concepts— is a bottleneck in many NLP / IE systems, usually conflated with typing, making it hard to see where models model span boundaries. ➡️"Entity-ness" should already be baked into LLM reps no ?

100 1

Victor Morand @victormorand.eurosky.social · 22/04/2026

🔍 Can we probe latent mention detection signals in Large Language Models ? Yes ! Excited to share that our paper ToMMeR (Token Matching for Mention Recognition) is heading to #ACL2026! We confirm that entity detection is a robust, latent capability in early LLM layers, and probe it ! 🧵

130 3

Reposted by Victor Morand

Sophia Sirko-Galouchenko 🇺🇦 @ssirko.bsky.social · 17/04/2026

1/n New paper - V-GIFT 🎁 Self-supervised tasks like rotation prediction or colorization were big in 2018. Do they still matter? Yes. We turn them into visual instruction tuning data for MLLMs. Result: models rely more on the image and perform better on vision tasks 👀

1248 24

Reposted by Victor Morand

Eurosky Social @eurosky.social · 09/03/2026

The future of social media is about you, not an app. Why we started by launching Eurosky accounts and our plans for apps. Read on from @sherif.eurosky.social

eurosky.leaflet.pub

A Eurosky Account is just the start

The future of social media is about you not an app

821492 214

Reposted by Victor Morand

Lê Nguyên Hoang @science4all.org · 06/03/2026

C'est la magie de l'interopérabilité granulaire du protocol AT de BlueSky. Plutôt que de déménager à avec valises et meubles, vous pouvez déménager juste vos valises, sans dépaysement. Et bientôt on vous aidera à déménager vos meubles aussi, toujours sans dépaysement. tournesol.app/entities/yt:...

tournesol.app

Quitter Twitter ? Mais pour aller où ?

Science4All

1102 10

Reposted by Victor Morand

Eurosky Social @eurosky.social · 06/03/2026

Almost 5k 🚀 let’s get there today - tell your friends! 🇫🇷 On passe les 5,000 aujourd’hui? Parlez d’Eurosky à vos amis! 🇪🇸 ¿Llegamos a 5,000 hoy? ¡Cuéntaselo a tus amigos! 🇩🇪 Schaffen wir heute 5,000? Erzählt euren Freunden von Eurosky! 🇳🇱 Halen we vandaag de 5,000? Vertel je vrienden over Eurosky!

11229123 229

Reposted by Victor Morand

David Bau @davidbau.bsky.social · 18/12/2025

I have been teaching myself to vibe code. Watch Claude Code grow my 780 lines to 13,600 - mandelbrot.page/coverage/ca... Two fundamental rules for staying in control: davidbau.com/archives/20...

2215 21

Victor Morand @victormorand.eurosky.social · 22/10/2025

👀Up next Building upon these findings, we've managed to externalize this internal mechanism, creating a general-purpose mention detector with promising results. Stay tuned! 🔜

000 0

Victor Morand @victormorand.eurosky.social · 22/10/2025

I'll be presenting this work at @blackboxnlp.bsky.social in Suzhou, happy to chat there or here if you are interested !

111 1

Victor Morand @victormorand.eurosky.social · 22/10/2025

3️⃣ The Entity Lens Our method enables reconstruction of entity mentions from any representation within LLMs, allowing to ask: “What entity is the model thinking about right now?” 💡 When reading ‘the City of Lights iconic monument’, the model internally “thinks” of Paris and the Eiffel Tower !

100 1

Victor Morand @victormorand.eurosky.social · 22/10/2025

2️⃣ LLMs develop entity-specific mechanisms. By sucessfully learning "Tasks Vectors" steering the model to reconstruct the mention, we uncover new evidence that LLMs form dedicated internal circuits to represent and manipulate multi-token entities.

100 1

Victor Morand @victormorand.eurosky.social · 22/10/2025

1️⃣ Common entities are (almost) part of the Vocabulary. We prove that common multi-token mentions (e.g. "Eiffel Tower") can be recovered from the middle-layer hidden state of its last token only ! Uncommon mentions aren't fully encoded this way; but rather retrieved from the context when needed.

100 1

Victor Morand @victormorand.eurosky.social · 22/10/2025

New paper at @blackboxnlp.bsky.social @ @emnlpmeeting.bsky.social ! ⚛️ Entities are the fundamental building blocks of knowledge. Although some clues emerge from mechanistic interpretability, how auto-regressive LLMs actually encode and retrieve them remains a mystery. 🧵 📄 arxiv.org/abs/2510.09421

111 1