MLIA ISIR
@mlia-isir.bsky.social
MLIA research team at CNRS/ISIR lab in Sorbonne University @sorbonne-universite.fr
https://www.isir.upmc.fr/equipes/mlia/
https://www.isir.upmc.fr/equipes/mlia/
Reposted by MLIA ISIR
🐭 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 🧵
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 🧵
September 1, 2026 at 11:25 AM
🐭 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 🧵
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 🧵
Last week, @victormorand.eurosky.social presented his work as a PhD student at MLIA's weekly seminar!
His presentation explored how transformers understand and represent named entities in their intermediate layers, and their implication for neural information retrieval.
Thanks, Victor!
His presentation explored how transformers understand and represent named entities in their intermediate layers, and their implication for neural information retrieval.
Thanks, Victor!
September 14, 2026 at 7:46 AM
Last week, @victormorand.eurosky.social presented his work as a PhD student at MLIA's weekly seminar!
His presentation explored how transformers understand and represent named entities in their intermediate layers, and their implication for neural information retrieval.
Thanks, Victor!
His presentation explored how transformers understand and represent named entities in their intermediate layers, and their implication for neural information retrieval.
Thanks, Victor!
At #SIGIR2026? Come say hi—we'd love to discuss our work!
📅 July 20-24, #SIGIR2026
📅 July 25th, #ICTIR2026
📍 Melbourne, Australia
📅 July 20-24, #SIGIR2026
📅 July 25th, #ICTIR2026
📍 Melbourne, Australia
July 23, 2026 at 10:05 AM
At #SIGIR2026? Come say hi—we'd love to discuss our work!
📅 July 20-24, #SIGIR2026
📅 July 25th, #ICTIR2026
📍 Melbourne, Australia
📅 July 20-24, #SIGIR2026
📅 July 25th, #ICTIR2026
📍 Melbourne, Australia
July 23, 2026 at 9:38 AM
We are very proud to share that the team was present at ACL:
1. ToMMeR – Efficient Entity Mention Detection from Large Language Models arxiv.org/abs/2510.19410
2. The GDN-CC Dataset: Automatic Corpus Clarification for AI-enhanced Democratic Citizen Consultations arxiv.org/abs/2601.14944
1. ToMMeR – Efficient Entity Mention Detection from Large Language Models arxiv.org/abs/2510.19410
2. The GDN-CC Dataset: Automatic Corpus Clarification for AI-enhanced Democratic Citizen Consultations arxiv.org/abs/2601.14944
The GDN-CC Dataset: Automatic Corpus Clarification for AI-enhanced Democratic Citizen Consultations
LLMs are ubiquitous in modern NLP, and while their applicability extends to texts produced for democratic activities such as online deliberations or large-scale citizen consultations, ethical question...
arxiv.org
July 9, 2026 at 10:01 AM
We are very proud to share that the team was present at ACL:
1. ToMMeR – Efficient Entity Mention Detection from Large Language Models arxiv.org/abs/2510.19410
2. The GDN-CC Dataset: Automatic Corpus Clarification for AI-enhanced Democratic Citizen Consultations arxiv.org/abs/2601.14944
1. ToMMeR – Efficient Entity Mention Detection from Large Language Models arxiv.org/abs/2510.19410
2. The GDN-CC Dataset: Automatic Corpus Clarification for AI-enhanced Democratic Citizen Consultations arxiv.org/abs/2601.14944
We are very proud to share that the team will be present at ICML this year again!
1. "PRISM: Perception Reasoning Interleaved for Sequential Decision Making". arxiv.org/abs/2605.05407
2. "Improving Zero-Shot Offline RL via Behavioral Task Sampling" arxiv.org/abs/2604.25496
1. "PRISM: Perception Reasoning Interleaved for Sequential Decision Making". arxiv.org/abs/2605.05407
2. "Improving Zero-Shot Offline RL via Behavioral Task Sampling" arxiv.org/abs/2604.25496
July 9, 2026 at 9:53 AM
We are very proud to share that the team will be present at ICML this year again!
1. "PRISM: Perception Reasoning Interleaved for Sequential Decision Making". arxiv.org/abs/2605.05407
2. "Improving Zero-Shot Offline RL via Behavioral Task Sampling" arxiv.org/abs/2604.25496
1. "PRISM: Perception Reasoning Interleaved for Sequential Decision Making". arxiv.org/abs/2605.05407
2. "Improving Zero-Shot Offline RL via Behavioral Task Sampling" arxiv.org/abs/2604.25496
🎉 Accepted at #CVPR2026 as oral in the EgoVis workshop 🎉
📄Action100M: A Large-scale Video Action Dataset
👥 D. Chen, T. Kasarla, Y. Bang, M. Shukor, W, Chung, J. Yu, A. Bolourchi, T. Moutakanni, P. Fung
📍 Denver, USA
📅 4 June
📖 arxiv.org/abs/2601.10592
Come listen to our presentation!
📄Action100M: A Large-scale Video Action Dataset
👥 D. Chen, T. Kasarla, Y. Bang, M. Shukor, W, Chung, J. Yu, A. Bolourchi, T. Moutakanni, P. Fung
📍 Denver, USA
📅 4 June
📖 arxiv.org/abs/2601.10592
Come listen to our presentation!
Action100M: A Large-scale Video Action Dataset
Inferring physical actions from visual observations is a fundamental capability for advancing machine intelligence in the physical world. Achieving this requires large-scale, open-vocabulary video act...
arxiv.org
June 24, 2026 at 12:31 PM
🎉 Accepted at #CVPR2026 as oral in the EgoVis workshop 🎉
📄Action100M: A Large-scale Video Action Dataset
👥 D. Chen, T. Kasarla, Y. Bang, M. Shukor, W, Chung, J. Yu, A. Bolourchi, T. Moutakanni, P. Fung
📍 Denver, USA
📅 4 June
📖 arxiv.org/abs/2601.10592
Come listen to our presentation!
📄Action100M: A Large-scale Video Action Dataset
👥 D. Chen, T. Kasarla, Y. Bang, M. Shukor, W, Chung, J. Yu, A. Bolourchi, T. Moutakanni, P. Fung
📍 Denver, USA
📅 4 June
📖 arxiv.org/abs/2601.10592
Come listen to our presentation!
This week we welcomed Arthur Ouaknine, Postdoc at
McGill University and Mila, to lead our team meeting.
He presented his recent work on “Remote sensing for forest monitoring: Learning to generalize at different scales”.
Thanks for the insightful presentation and discussions!
McGill University and Mila, to lead our team meeting.
He presented his recent work on “Remote sensing for forest monitoring: Learning to generalize at different scales”.
Thanks for the insightful presentation and discussions!
June 22, 2026 at 11:08 AM
This week we welcomed Arthur Ouaknine, Postdoc at
McGill University and Mila, to lead our team meeting.
He presented his recent work on “Remote sensing for forest monitoring: Learning to generalize at different scales”.
Thanks for the insightful presentation and discussions!
McGill University and Mila, to lead our team meeting.
He presented his recent work on “Remote sensing for forest monitoring: Learning to generalize at different scales”.
Thanks for the insightful presentation and discussions!
📢 New PhD defense ! 📢
This week, Lise Le Boudec will defend her thesis entitled:
"Meshless Neural Solvers and Generative Operators for Parametric PDEs"
📅 Wednesday, June 10
🕒 2:00 PM
📍Amphitheater 45A, 4 place Jussieu, Paris
The defense is open to the public and will be broadcasted live 📡
This week, Lise Le Boudec will defend her thesis entitled:
"Meshless Neural Solvers and Generative Operators for Parametric PDEs"
📅 Wednesday, June 10
🕒 2:00 PM
📍Amphitheater 45A, 4 place Jussieu, Paris
The defense is open to the public and will be broadcasted live 📡
June 8, 2026 at 12:37 PM
📢 New PhD defense ! 📢
This week, Lise Le Boudec will defend her thesis entitled:
"Meshless Neural Solvers and Generative Operators for Parametric PDEs"
📅 Wednesday, June 10
🕒 2:00 PM
📍Amphitheater 45A, 4 place Jussieu, Paris
The defense is open to the public and will be broadcasted live 📡
This week, Lise Le Boudec will defend her thesis entitled:
"Meshless Neural Solvers and Generative Operators for Parametric PDEs"
📅 Wednesday, June 10
🕒 2:00 PM
📍Amphitheater 45A, 4 place Jussieu, Paris
The defense is open to the public and will be broadcasted live 📡
🎉 Accepted at #ICRA2026 🎉
The MLIA team will present:
📄I-FailSense: Towards General Robotic Failure Detection with Vision-Language Models
👥 Clémence Grislain, Hamed Rahimi, Olivier Sigaud, Mohamed Chetouani
📍 Vienna, Austria
📅 June 1–5
📖 arxiv.org/abs/2509.16072
Come say hi at our poster session!
The MLIA team will present:
📄I-FailSense: Towards General Robotic Failure Detection with Vision-Language Models
👥 Clémence Grislain, Hamed Rahimi, Olivier Sigaud, Mohamed Chetouani
📍 Vienna, Austria
📅 June 1–5
📖 arxiv.org/abs/2509.16072
Come say hi at our poster session!
I-FailSense: Towards General Robotic Failure Detection with Vision-Language Models
Language-conditioned robotic manipulation in open-world settings requires not only accurate task execution but also the ability to detect failures for robust deployment in real-world environments. Alt...
arxiv.org
June 2, 2026 at 10:20 AM
🎉 Accepted at #ICRA2026 🎉
The MLIA team will present:
📄I-FailSense: Towards General Robotic Failure Detection with Vision-Language Models
👥 Clémence Grislain, Hamed Rahimi, Olivier Sigaud, Mohamed Chetouani
📍 Vienna, Austria
📅 June 1–5
📖 arxiv.org/abs/2509.16072
Come say hi at our poster session!
The MLIA team will present:
📄I-FailSense: Towards General Robotic Failure Detection with Vision-Language Models
👥 Clémence Grislain, Hamed Rahimi, Olivier Sigaud, Mohamed Chetouani
📍 Vienna, Austria
📅 June 1–5
📖 arxiv.org/abs/2509.16072
Come say hi at our poster session!
🎓️ PhD defense coming!
Armand Kassaï Koupaï will defend his thesis:
"Scalable Neural Operators for Parametric PDEs: Adaptive and Generative Methods" on Wed. June 3rd 📅
⌚️️3pm
📍Amphi Herpin, Bat. ESCLANGON, 4 place Jussieu, Paris
The defense will be broadcast live.
Armand Kassaï Koupaï will defend his thesis:
"Scalable Neural Operators for Parametric PDEs: Adaptive and Generative Methods" on Wed. June 3rd 📅
⌚️️3pm
📍Amphi Herpin, Bat. ESCLANGON, 4 place Jussieu, Paris
The defense will be broadcast live.
June 2, 2026 at 9:38 AM
🎓️ PhD defense coming!
Armand Kassaï Koupaï will defend his thesis:
"Scalable Neural Operators for Parametric PDEs: Adaptive and Generative Methods" on Wed. June 3rd 📅
⌚️️3pm
📍Amphi Herpin, Bat. ESCLANGON, 4 place Jussieu, Paris
The defense will be broadcast live.
Armand Kassaï Koupaï will defend his thesis:
"Scalable Neural Operators for Parametric PDEs: Adaptive and Generative Methods" on Wed. June 3rd 📅
⌚️️3pm
📍Amphi Herpin, Bat. ESCLANGON, 4 place Jussieu, Paris
The defense will be broadcast live.
We were delighted to welcome Tanise Ceron @taniseceron.bsky.social (MilaNLP, Bocconi University) to lead our team meeting. She shared her latest research on
"Political Biases in LLMs: Evidence from Training Data and Model Behavior."
Many thanks for the insightful presentation and discussion👏💡
"Political Biases in LLMs: Evidence from Training Data and Model Behavior."
Many thanks for the insightful presentation and discussion👏💡
June 1, 2026 at 12:39 PM
We were delighted to welcome Tanise Ceron @taniseceron.bsky.social (MilaNLP, Bocconi University) to lead our team meeting. She shared her latest research on
"Political Biases in LLMs: Evidence from Training Data and Model Behavior."
Many thanks for the insightful presentation and discussion👏💡
"Political Biases in LLMs: Evidence from Training Data and Model Behavior."
Many thanks for the insightful presentation and discussion👏💡
On ISIR Young Scientists Day, Guillaume Jeanneret presented FrescoDiffusion for 4K image‑to‑video generation, and @palequeu.bsky.social presented the GDN‑CC Dataset for improved corpus clarification in democratic consultations, alongside team members showcasing their work in the poster session.
May 26, 2026 at 2:43 PM
On ISIR Young Scientists Day, Guillaume Jeanneret presented FrescoDiffusion for 4K image‑to‑video generation, and @palequeu.bsky.social presented the GDN‑CC Dataset for improved corpus clarification in democratic consultations, alongside team members showcasing their work in the poster session.
🎉 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
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
April 24, 2026 at 5:44 PM
🎉 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
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
🎓️ We’re thrilled to announce that Lucas SCHOTT will soon defend his PhD thesis!
📖Title: "Diverse and Adaptive Adversarial Training for Robust Reinforcement Learning".
📅Tuesday, 14th April 2026
⌚️️2pm
📍Amphi Astier, Batiment Esclangon, 4 Place Jussieu, Paris
#PhD #RL #PhdDefense #agent
📖Title: "Diverse and Adaptive Adversarial Training for Robust Reinforcement Learning".
📅Tuesday, 14th April 2026
⌚️️2pm
📍Amphi Astier, Batiment Esclangon, 4 Place Jussieu, Paris
#PhD #RL #PhdDefense #agent
April 10, 2026 at 9:27 AM
🎓️ We’re thrilled to announce that Lucas SCHOTT will soon defend his PhD thesis!
📖Title: "Diverse and Adaptive Adversarial Training for Robust Reinforcement Learning".
📅Tuesday, 14th April 2026
⌚️️2pm
📍Amphi Astier, Batiment Esclangon, 4 Place Jussieu, Paris
#PhD #RL #PhdDefense #agent
📖Title: "Diverse and Adaptive Adversarial Training for Robust Reinforcement Learning".
📅Tuesday, 14th April 2026
⌚️️2pm
📍Amphi Astier, Batiment Esclangon, 4 Place Jussieu, Paris
#PhD #RL #PhdDefense #agent
Reposted by MLIA ISIR
We are very happy to announce our next seminar: Paul Lerner @lernerp.bsky.social (ISIR, Sorbonne Université & CNRS) "Controlling Linguistic Variability in Large Language Models" on Friday 10th April 2026, 11am CET. Details here 👉 almanach.inria.fr/seminars-en....
April 3, 2026 at 9:07 AM
We are very happy to announce our next seminar: Paul Lerner @lernerp.bsky.social (ISIR, Sorbonne Université & CNRS) "Controlling Linguistic Variability in Large Language Models" on Friday 10th April 2026, 11am CET. Details here 👉 almanach.inria.fr/seminars-en....
Accepted papers at EACL 2026 🎉
Several members from the team are in Rabat, Morocco, this week to present their work! Don't hesitate to drop by and have a chat with them 🙋
List in thread ⬇️
Several members from the team are in Rabat, Morocco, this week to present their work! Don't hesitate to drop by and have a chat with them 🙋
List in thread ⬇️
March 24, 2026 at 10:09 AM
Accepted papers at EACL 2026 🎉
Several members from the team are in Rabat, Morocco, this week to present their work! Don't hesitate to drop by and have a chat with them 🙋
List in thread ⬇️
Several members from the team are in Rabat, Morocco, this week to present their work! Don't hesitate to drop by and have a chat with them 🙋
List in thread ⬇️
🎓Wrapping up March on a high note with the thesis defense of Tanguy Herserant!
📖Title: "Amélioration des métriques d'évaluation pour les méthodes de résumé automatique".
📅Monday, 30th March
⌚️2pm
📍Amphi Herpin, Bat. ESCLANGON
📖Title: "Amélioration des métriques d'évaluation pour les méthodes de résumé automatique".
📅Monday, 30th March
⌚️2pm
📍Amphi Herpin, Bat. ESCLANGON
March 24, 2026 at 9:44 AM
🎓Wrapping up March on a high note with the thesis defense of Tanguy Herserant!
📖Title: "Amélioration des métriques d'évaluation pour les méthodes de résumé automatique".
📅Monday, 30th March
⌚️2pm
📍Amphi Herpin, Bat. ESCLANGON
📖Title: "Amélioration des métriques d'évaluation pour les méthodes de résumé automatique".
📅Monday, 30th March
⌚️2pm
📍Amphi Herpin, Bat. ESCLANGON
Loïck Chambon is defending his PhD thesis titled "Improving Representations for Autonomous Driving". This PhD was conducted in collaboration with @valeoai.bsky.social.
The presentation will start today at 15:15. 🎉
Follow remotely ⬇️
www.youtube.com/live/uFMAHn0...
The presentation will start today at 15:15. 🎉
Follow remotely ⬇️
www.youtube.com/live/uFMAHn0...
PhD defense - Loïck CHAMBON
YouTube video by mlia deeplearning
www.youtube.com
February 23, 2026 at 2:14 PM
Loïck Chambon is defending his PhD thesis titled "Improving Representations for Autonomous Driving". This PhD was conducted in collaboration with @valeoai.bsky.social.
The presentation will start today at 15:15. 🎉
Follow remotely ⬇️
www.youtube.com/live/uFMAHn0...
The presentation will start today at 15:15. 🎉
Follow remotely ⬇️
www.youtube.com/live/uFMAHn0...
This week we welcomed Mingmeng Geng @mingmenggeng.bsky.social, Postdoc at @normalesup.bsky.social & @labolattice.bsky.social, to lead our team meeting.
He presented his recent work on “Analysis and Detection of LLM-Generated Content”.
Thanks for the insightful presentation and discussions!
He presented his recent work on “Analysis and Detection of LLM-Generated Content”.
Thanks for the insightful presentation and discussions!
February 9, 2026 at 1:47 PM
This week we welcomed Mingmeng Geng @mingmenggeng.bsky.social, Postdoc at @normalesup.bsky.social & @labolattice.bsky.social, to lead our team meeting.
He presented his recent work on “Analysis and Detection of LLM-Generated Content”.
Thanks for the insightful presentation and discussions!
He presented his recent work on “Analysis and Detection of LLM-Generated Content”.
Thanks for the insightful presentation and discussions!
Today was the "Galette des rois" or Epiphany Cake for the team!
A great occasion to meet together for a friendly moment! 👑
A great occasion to meet together for a friendly moment! 👑
January 19, 2026 at 4:25 PM
Today was the "Galette des rois" or Epiphany Cake for the team!
A great occasion to meet together for a friendly moment! 👑
A great occasion to meet together for a friendly moment! 👑
First team meeting of 2026 🚀 ✅
Sophia Sirko @ssirko.bsky.social presented her #ICCV2025 paper:
"DIP: Unsupervised Dense In-Context Post-training of Visual Representations"
Thanks for the insightful presentation! ! 👏
Sophia Sirko @ssirko.bsky.social presented her #ICCV2025 paper:
"DIP: Unsupervised Dense In-Context Post-training of Visual Representations"
Thanks for the insightful presentation! ! 👏
January 12, 2026 at 1:37 PM
First team meeting of 2026 🚀 ✅
Sophia Sirko @ssirko.bsky.social presented her #ICCV2025 paper:
"DIP: Unsupervised Dense In-Context Post-training of Visual Representations"
Thanks for the insightful presentation! ! 👏
Sophia Sirko @ssirko.bsky.social presented her #ICCV2025 paper:
"DIP: Unsupervised Dense In-Context Post-training of Visual Representations"
Thanks for the insightful presentation! ! 👏
📢Offre de stage de M2, sous la tutelle d'Olivier Schwander:
Je cherche un/une stagiaire de M2 informatique ou mathématique (spécialité science des données, apprentissage statistique et intelligence artificielle): *Détection et caractérisation non-supervisée d’arbres à partir d’images aériennes* ⬇️
Je cherche un/une stagiaire de M2 informatique ou mathématique (spécialité science des données, apprentissage statistique et intelligence artificielle): *Détection et caractérisation non-supervisée d’arbres à partir d’images aériennes* ⬇️
December 5, 2025 at 4:30 PM
📢Offre de stage de M2, sous la tutelle d'Olivier Schwander:
Je cherche un/une stagiaire de M2 informatique ou mathématique (spécialité science des données, apprentissage statistique et intelligence artificielle): *Détection et caractérisation non-supervisée d’arbres à partir d’images aériennes* ⬇️
Je cherche un/une stagiaire de M2 informatique ou mathématique (spécialité science des données, apprentissage statistique et intelligence artificielle): *Détection et caractérisation non-supervisée d’arbres à partir d’images aériennes* ⬇️
Congratulations Loïck Chambon, Paul Couairon, Nicolas Thome, Matthieu Cord and the @valeoai.bsky.social team!
Paper: arxiv.org/abs/2511.18452
by Loïck Chambon (loickch.github.io), @paulcouairon.bsky.social, @eloizablocki.bsky.social, @alexandreboulch.bsky.social, @nicolasthome.bsky.social, @matthieucord.bsky.social
Collab with @mlia-isir.bsky.social
by Loïck Chambon (loickch.github.io), @paulcouairon.bsky.social, @eloizablocki.bsky.social, @alexandreboulch.bsky.social, @nicolasthome.bsky.social, @matthieucord.bsky.social
Collab with @mlia-isir.bsky.social
NAF: Zero-Shot Feature Upsampling via Neighborhood Attention Filtering
Vision Foundation Models (VFMs) extract spatially downsampled representations, posing challenges for pixel-level tasks. Existing upsampling approaches face a fundamental trade-off: classical filters a...
arxiv.org
November 29, 2025 at 4:59 PM
Congratulations Loïck Chambon, Paul Couairon, Nicolas Thome, Matthieu Cord and the @valeoai.bsky.social team!