A kind reminder that the deadline for the ICLR2025 Blogpost track is Nov. 15th, 23:59AOE, 2024. If you have any questions, please email us at: [email protected] #ICLR2025 ICLR 2026

We are extending the submission deadline by one week to the 22nd of Nov. Looking forward to your submissions!

📣🔥Last chance to submit your work for the ICLR 2026 2025 Blogpost Track! The deadline is November 22, 2024, at 23:59 AOE—don’t miss it! For any inquiries, feel free to reach out to us at: [email protected]. iclr.cc/Conferences/20… #MachineLearning

I'll be hiring PhD and MSc students for Fall 2025 to do research on machine learning for science (details in 🧵) I especially encourage applicants from under-represented groups and regions in conflict (Palestine, Ukraine, Sudan, Myanmar...) See 🧵 for how to apply.

We provide a friendly and visual introduction to flow matching techniques! w. Anne Gagneux, Ségolène Martin, Mathurin Massias , and Rémi Emonet (the king for figures)

I will be at NeurIPS! Would love to chat about research! Especially about fine-tuning of LLMs, generative models more generally, and reasoning! I will be presenting "Efficient Adversarial Training in LLMs with Continuous Attacks" (spotlight) at the poster session on Thursday!

Here is a one-minute summary of Sophie Xhonneux (Sophie Xhonneux)'s Efficient Adversarial Training in LLMs with Continuous Attacks. Come see the spotlight poster at NeurIPS Conference today: Poster Session 3 East, #4702

Come to our poster #4702! East Exhibit Halls A-C

Safe and reliable machine learning has never been more relevant. However, at the same time, LLMs made robustness research **more complex, less reproducible, and harder-to-evaluate**. How can we enable research progress despite these challenges?🧵 Stephan Günnemann Gauthier Gidel

We hope that our position will help to establish useful guidelines in the community and invite other researchers to do a retrospective on our claims next year! 🔗arxiv.org/abs/2502.11910 Yan Scholten Tom Wollschläger Sophie Xhonneux Cas (Stephen Casper) Stephan Günnemann Gauthier Gidel

Leo Schwinn led a new position paper discussing how LLM robustness research is becoming harder! It includes many suggestions that can help us make some progress in the field.

mrinank 🍂 Very exciting work! We released a paper on query-efficient ways to predict worst-case risks a few days ago: arxiv.org/abs/2502.10487 Our scaling curves end a lot earlier though ;-)

Our work on Asynchronous RLHF was accepted to #ICLR2025 ! (I was so excited to announce it, I forgot to say I was excited) Used by Ai2 for OLMo-2 32B 🔥 New results show ~70% speedups for LLM + RL math and reasoning 🧠 🧵below or hear my DLCT talk online on March 28!

I am at ICLR this year, please reach out if you would like to have a chat.

Mark your calendars! The @WiMLWorkshop Social returns to ICLR 2026! 📅 🗓️ Friday, April 25th, 2025 ⏰ 12:30 PM - 2:00 PM 📍 Hall 1 Apex, Singapore Expo iclr.cc/virtual/2025/s… lu.ma/ftdnjnf9 #ICLR2025

Join us for the WiML Social at #ICLR2025 in ~2.5 hours! Vasiliki (Vasiliki Tassopoulou) and Melis (Melis İlayda Bal) have organized a fun event with: icebreaker game w/ a prize, topic roundtables, and a fantastic panel of women in AI/ML who have joint academic+industry experience 💙

If you are at ICLR 2026 and are interested in making your RLHF really fast come find Michael Noukhovitch, used to be @ICLR 2025 and me at poster #582.

Following a successful pilot, ICLR has decided to continue its partnership with TMLR. Look forward to seeing more exciting TMLR publications featured at ICLR for the foreseeable future!

How do LLMs navigate refusal? Our new @ICMLConf paper introduces a gradient-based approach & Representational Independence to map this complex internal geometry. 🚨 New Research Thread! 🚨 The Geometry of Refusal in Large Language Models By Stephan Günnemann's lab & Google AI. 🧵👇

It's very difficult to improve the *exponent* in scaling laws for loss vs compute, especially by changing the optimizer! Our new paper shows that scaling momentum correctly can *provably* improve the scaling exponent on a theoretical model. Empirically, it works on LSTMs too!