@mlia-isir.bsky.social
MLIA research team at CNRS/ISIR lab in Sorbonne University @sorbonne-universite.fr https://www.isir.upmc.fr/equipes/mlia/
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
Accepted to a Workshop (2/2):
"On the Representations of Entities in Auto-regressive Large Language Models", accepted to the BlackBoxNLP Workshop, from @victormorand.bsky.social, Josiane Mothe & Benjamin Piwowarski
🎤 Nov 9th, 2025
Arxiv: arxiv.org/abs/2510.09421
Code: github.com/VictorMorand...
Accepted to a Workshop (1/2):
"Self-Retrieval from Distant Contexts for Document-Level Machine Translation", accepted to the Conference on Machine Translation (WMT25), from @ziqianpeng.bsky.social, @rachelbawden.bsky.social, @yvofr.bsky.social
Accepted to the Findings of EMNLP (3/3):
"How Sampling Affects the Detectability of Machine-written Texts: A Comprehensive Study", from Matthieu Dubois, François Yvon, Pablo Piantanida.
🎤 Wed. Nov 5 at 13:00-14:00
arxiv.org/abs/2510.13681
In collaboration with BNP Paribas.
Arxiv: arxiv.org/abs/2509.10663
Code: github.com/Zineddine-Ti...
Project page: zineddine-tighidet.github.io/projects/pro...
Accepted to the Findings of EMNLP (2/3):
"Context Copying Modulation: The Role of Entropy Neurons in Managing Parametric and Contextual Knowledge Conflicts" by Zineddine Tighidet, Andrea Mogini, Hedi Ben-younes, Jiali Mei, Patrick Gallinari, Benjamin Piwowarski.
🎤Thu. Nov 6 at 12:30-13:30 Hall C
Accepted to the Findings of EMNLP (1/3):
"Tracing Multilingual Factual Knowledge Acquisition in Pretraining" from Yihong Liu, Mingyang Wang, Amir Hossein Kargaran, Felicia Körner, Ercong Nie, Barbara Plank, @yvofr.bsky.social and Hinrich Schütze, with LMU Munich.
arxiv.org/abs/2505.14824
Authors: M. Carpuat, O. Asscher, K. Bali, L. Bentivogli, F. Blain, L. Bowker, M. Choudhury, H. Daumé III, K. Duh, G. Gao, A. Grissom II, M. Karpinska, E. C. Khoong, W. D. Lewis, A. F. T. Martins, M. Nurminen, D. W. Oard, M. Popovic, M. Simard and @yvofr.bsky.social
28.10.2025 08:57 — 👍 0 🔁 0 💬 1 📌 0Accepted to the Main Conference (2/2):
"An Interdisciplinary Approach to Human-Centered Machine Translation", in collaboration with the University of Maryland.
(list of authors thereafter ➡️)
arxiv.org/abs/2506.13468
Accepted to the Main Conference (1/2):
"Relation-Specific Neurons in Large Language Models", from Yihong Liu, Runsheng Chen, Lea Hirlimann, Ahmad Dawar Hakimi, Mingyang Wang, Amir Hossein Kargaran, Sascha Rothe, @yvofr.bsky.social and Hinrich Schütze, with LMU Munich.
arxiv.org/abs/2502.17355
Hope your tickets to attend @emnlpmeeting.bsky.social 2025 are booked! 🛫🇨🇳
We are very proud to share that the team will be present this year again, with nothing less than 7 accepted papers! 🎉
📍Suzhou, China
📅Nov 4 - 9th, 2025
Details and papers ⬇️
🎉 PhD defense announcement!
@paulcouairon.bsky.social will defend his PhD titled "Structured Visual Representations with Foundation Models"!
🗓️When? Thursday, 4th November at 14:00
📍Where? Room 304 65/66, campus Pierre et Marie Curie, Sorbonne Université (4 place Jussieu, 75005 Paris).
Abstract ⬇️
During today's team meeting, Marc Lafon from @lecnam.bsky.social presented his recent paper, accepted at @iccv.bsky.social 🏝️: "Learning vision-language uncertainties for failure prediction" (arxiv.org/abs/2507.07620)
13.10.2025 11:52 — 👍 0 🔁 0 💬 0 📌 0"GaussRender: Learning 3D Occupancy with Gaussian Rendering" from Loïck Chambon, Eloi Zablocki , Alexandre BOULCH, Mickael Chen, Matthieu Cord, in collaboration with @valeoai.bsky.social.
Paper and code ⬇️
📑 arxiv.org/abs/2502.05040
🖥️ github.com/valeoai/Gaus...
"Analyzing Fine-tuning Representation Shift for Multimodal LLMs Steering alignment" from pegah khayatan, Mustafa Shukor, Jayneel Parekh*, arnaud dapogny, Matthieu Cord.
Paper, code and project page: ⬇️
📑 arxiv.org/pdf/2501.03012
🖥️ github.com/mshukor/xl-v...
➡️ pegah-kh.github.io/projects/lmm...
"DIP: Unsupervised Dense In-Context Post-training of Visual Representations" from @ssirko.bsky.social , @spyrosgidaris.bsky.social, @vobeckya.bsky.social, @abursuc.bsky.social and Nicolas Thome, with @valeoai.bsky.social.
Paper and code ⬇️
📑 arxiv.org/abs/2506.18463
🖥️ github.com/sirkosophia/...
💫 Accepted paper at @iccv.bsky.social
The team is proud to announce that 3 papers were accepted to #ICCV2025! 🥳
📍 Honolulu, Hawaï, USA
🗓️ Oct 19 – 23th, 2025
Looking forward to present our works and discuss about it! 💫
Details and papers ⬇️
Accepted paper at IROS 2025 🥳
🗓️ 19-25th October 2025
📍 Hangzhou, China
Zakariae EL ASRI, Ibrahim Laiche, Clément Rambour, Olivier Sigaud, Nicolas Thome
"RT-HCP: Dealing with Inference Delays and Sample Efficiency to Learn Directly on Robotic Platforms"
➡️ See the GitHub at github.com/elasriz/RTHCP
Holidays are already over (sigh) but team's meetings are back for another year! 🔬
Today, we're listening to Jayneel Parekh, post-doc student under the supervision of @matthieucord.bsky.social , talking about his recent work: "Learning to Steer: Input-dependent Steering for Multimodal LLMs"
📝 arxiv.org/abs/2508.20312
💻 github.com/BenKabongo25...
@recsys.bsky.social is happening this week in Prague, Czech Republic, and Ben Kabongo will be there to present his paper, together with Vincent Guigue and Pirmin Lemberger from OnePoint 🚀
➡️"ELIXIR: Efficient and LIghtweight model for eXplaIning Recommendations"
📆 EARL Workshop
📸 Some snapshots from the #ACL2025 conference last week!
Bravo for the MLIA people! Thanks a lot to all the insightful discussions!
Recap' of the team's activities
bsky.app/profile/mlia...
#NLP #ACL
The defense is starting!
If you cannot attend in person, follow online! ⬇️
www.youtube.com/live/CYGxEin...
- "Improving Diversity in Language Models: When Temperature Fails, Change the Loss" from Alexandre Verine, Florian Le Bronnec, Kunhao Zheng, Alexandre Allauzen, Yann Chevaleyre, Benjamin Negrevergne, with Miles lab from PSL
📄 openreview.net/forum?id=Rsy...
👤 Poster: Wed 16 Jul 11 a.m. - 1:30 p.m.
- "Zebra: In-Context Generative Pretraining for Solving Parametric PDEs" from Louis Serrano, @armandkassai.bsky.social, Thomas Wang, Pierre ERBACHER, Patrick Gallinari
📄 arxiv.org/abs/2410.03437
🖥️ github.com/LouisSerrano...
👤 Poster: Thu 17 Jul 11 a.m. PDT — 1:30 p.m. PDT
- "MAGELLAN: Metacognitive predictions of learning progress guide autotelic LLM agents in large goal spaces" from Loris Gaven, Thomas Carta, Clément Romac, Cédric Colas, Sylvain Lamprier, Olivier Sigaud, Pierre-Yves Oudeyer
📄 arxiv.org/abs/2502.07709
👤 Poster: Thu 17 Jul 11 a.m. PDT — 1:30 p.m. PDT
- "DISCO: learning to DISCover an evolution Operator for multi-physics-agnostic prediction" from @rdmorel.bsky.social, @jiequnh.bsky.social, Edouard Oyallon, in collaboration with #FlatironCCM.
📄 arxiv.org/abs/2504.19496
👤 Poster: Wed 16 Jul 11 a.m. PDT — 1:30 p.m. PDT
📣 Accepted paper at ICML 2025 📣
The team is proud to annonce that 4 papers were accepted at #ICML2025.
📍Vancouver, Canada
🗓️ 13th - 19th July
Looking forward to present our works and discuss about it!
Papers and schedule following ⬇️
CORIA-TALN recap'! ☀️
Last week, the team was in Marseille to attend to the CORIA-TALN conference. The team was presenting several posters, workshops and oral presentations!
💡 Interested in participating to the MaTOS ANR's experiment?
➡️ anr-matos.github.io/posts/2025/j...
🎉 PhD defense announcement!
Song Duong will defend his PhD "Faithful and Efficient Conditional Text Generation for Data-to-text and Summarization"!
🗓️ Thursday, 10th July at 14:00
📍 Amphithéâtre Herpin, Bâtiment Esclangon, 4 place Jussieu, 75005 Paris.
Abstract ⬇️
