Found on the site which must not be named. #ICLR2025

PolyNet Boosts Exploration in Neural Combinatorial Optimization

PolyNet introduces a single‑decoder RL model that learns diverse strategies, boosting exploration without rules. Tested on four combinatorial optimization tasks, it was accepted at ICLR 2025. getnews.me/polynet-boosts-explorati... #polynet #iclr2025

𝗠𝗖𝗠𝗟 𝗕𝗹𝗼𝗴: Ever rerun feature importance plots and get different “top features”? That randomness comes from how explanations sample data.

A #ICLR2025 paper by MCML researchers flips this: Compress Then Explain (CTE) makes model explanations more accurate.

📄 Read more: mcml.ai/news/2025-09...

T2V‑Turbo‑v2 Enhances Text‑to‑Video Generation via Data and Rewards

T2V-Turbo-v2, using post-training consistency distillation with high-quality data and reward-model guidance, scored 85.13 on VBench and was accepted for presentation at ICLR 2025. Read more: getnews.me/t2v-turbo-v2-enhances-te... #t2vturbo #vbench #iclr2025

Vector researchers dive into deep learning at ICLR 2025 - Vector Institute for Artificial Intelligence Vector researchers made significant contributions to this year’s International Conference on Learning Representations (ICLR), the world’s leading venue for representation learning and deep learning re...

ICYMI - Vector researchers made a major impact at #ICLR2025 with 71 accepted papers!

Discover the cutting-edge research shaping AI's future 👇

vectorinstitute.ai/vector-resea...

Vector researchers dive into deep learning at ICLR 2025 - Vector Institute for Artificial Intelligence Vector researchers made significant contributions to this year’s International Conference on Learning Representations (ICLR), the world’s leading venue for representation learning and deep learning re...

Au cas où vous l'auriez manqué : les chercheurs de Vecteur ont fait forte impression lors de la conférence #ICLR2025 avec 71 articles acceptés !

Découvrez les recherches de pointe qui façonnent l'avenir de l'IA 👇

vectorinstitute.ai/vector-resea...

Check out our MOCA self-supervised learning approach unifying the learning principles of both discriminative & masked image modelling paradigms.
With a non-linear path, MOCA has been accepted at #TMLR and presented in the TMLR poster session at #iclr2025

12/
- we tried again, but not enough bold numbers 😞
- by then DINOv2 was released & the world has changed (for the good)
- MOCA eventually landed at #TMLR (top reviewers) & eventually presented as poster at #iclr2025

*photo: running downstream exps during rebuttal at judo match

1/ New & old work on self-supervised representation learning (SSL) with ViTs:
MOCA ☕ - Predicting Masked Online Codebook Assignments w/ @spyrosgidaris.bsky.social O. Simeoni, A. Vobecky, @matthieucord.bsky.social, N. Komodakis, @ptrkprz.bsky.social #TMLR #ICLR2025
Grab a ☕ & brace for a story & a🧵

🚀 We’ve released the source code for 𝗔𝗦𝗜𝗗𝗘 (presented as an 𝗢𝗿𝗮𝗹 at the #ICLR2025 BuildTrust workshop)!

🔍 ASIDE boosts prompt injection robustness without safety-tuning: we simply rotate embeddings of marked tokens by 90° during instruction-tuning and inference.

👇 code & docs👇

Had an awesome time presenting mine and @jfoerst.bsky.social's "AgentBreeder" paper at #ICLR2025 to a room of over a hundred researchers! Our approach addresses pre-deployment Al safety risks in multi-agent systems by evolving thousands of "scaffolds" from base LLMs in a multi-objective setting.

Have you had a paper accepted to #ACL2025 or #ICML2025, or did you present at #ICLR2025? Join us at @kingscollegelondon.bsky.social on 25 June for PreTrain, building a vibrant network of #ML and #NLP scholars.

Submit proposals by 13 June and register by 18 June pretrain2025.github.io/index.html

1/ Controlling LLMs with steering vectors is unreliable, but why?  Our paper, "Understanding (Un)Reliability of Steering Vectors in Language Models," at the #ICLR2025 Workshop on
Foundation Models in the Wild investigates this! What did we find?

Learning at peak efficiency Optimizing transport, trust, and tutelage.

The Data Science and Machine Learning Unit at OIST is working to improve the algorithms of the information age and how we interact with AI💻⚙ They had 5 papers – 4 written by interns – accepted at the recent #ICLR2025. Learn more about the unit 👇 www.oist.jp/news-center/...

What an unforgettable experience at #ICLR2025! 🎉

Still soaking in everything I learned, the inspiring conversations I had, and the amazing connections I made over the past few days.

1/4 🚀 New #ICLR2025 SPOTLIGHT ALERT
Gerstein Lab presents “Enhancing the Scalability & Applicability of Kohn-Sham Hamiltonians”—led by YunyangLI & Z Xia & L Huang & J Zhang & @markgerstein.bsky.social . Joint work with @msftresearch.bsky.social .

Excited to announce our #ICLR2025 spotlight work deriving the first exact certificates for neural networks against label poisoning 🎉. Joint work with @maha-saba.bsky.social, Stephan Günnemann & Debarghya Ghoshdastidar. For details, check out the thread below👇 or our paper arxiv.org/abs/2412.00537.

#ICLR2025 recap' 🇸🇬
Great time at ICLR2025 for our team, who presented their work during the poster sessions!
➡️ SCOPE: A Self-supervised Framework for Improving Faithfulness in Conditional Text Generation iclr.cc/virtual/2025...

Equivariance is dead - long live equivariance? Scaling pure Transformers is the future of molecular generative models?

I think its a lot more nuanced -- I tried to give some perspectives in my spotlight talk at #ICLR2025 AI4Mat Workshop last week

Full video here: youtu.be/NiY4NLzemnU?...

Uno studio della Johns Hopkins University dimostra che l’intelligenza artificiale non riesce a interpretare le interazioni sociali dinamiche come fa l’essere umano

#guidaautonoma #ICLR2025 #IntelligenzaArtificiale #interazionesociale #JohnsHopkinsUniversity
www.matricedigitale.it/tech/robotic...

#ICYMI the ESP research team presented at #ICLR2025 last week, sharing the latest updates on NatureLM-audio.

Thanks to everyone who stopped by to connect, exchange ideas & hear about this exciting work.

Explore NatureLM-audio, now open-sourced:
earthspecies.github.io/naturelm-aud...

It was a pleasure to sponsor the Bidirectional Human-AI Alignment workshop at #ICLR2025 Singapore and #CHI2025 Japan this week.

The workshop focused on the bidirectional nature of alignment—not just shaping AI for humans, but empowering humans to understand and work with AI. #AcademicSky

Tips on How to Connect at Academic Conferences I was a kinda awkward teenager. If you are a CS researcher reading this post, then chances are, you were too. How to navigate social situations and make friends is not always intuitive, and has to …

I wrote a post on how to connect with people (i.e., make friends) at CS conferences. These events can be intimidating so here's some suggestions on how to navigate them

I'm late for #ICLR2025 #NAACL2025, but in time for #AISTATS2025 #ICML2025! 1/3
kamathematics.wordpress.com/2025/05/01/t...

It's a wrap! 🎉 Thank you @iclr-conf.bsky.social for supporting so many inspiring ideas and thanks Singapore for having me again! 🇸🇬 #ICLR2025 #MachineLearning #PhDlife

Thanks to everyone involved in the AI for Nucleic Acids Workshop at #ICLR2025! And oh my god - the panel was a 🌶️🌶️🌶️ finish to an exciting day of science, discussions, and community!

@pranam.bsky.social @kishwar.bsky.social @bunnech.bsky.social and many others!

Last week, our PhD student @brtrm.bsky.social presented "Improving Reasoning Performance in Large Language Models via Representation Engineering" at ICLR 2025 🤖💭

If you missed his presentation, check out the paper here arxiv.org/abs/2504.19483

#NLProc #ICLR2025

#ICLR2025 & #ACMHCI are a wrap! 🎉 Couldn't ask for better vibes, conversations, and food.

🧬 BirdieDNA
🗣️ SLP Sidekick
🤖 Agents for Genomics

Big thanks to my co-authors @defnecirci.bsky.social, Flavia Negrete, Celeste Watkins, Asher Moldwin & Amarda Shehu, absolute rockstars 🤝🙏

@ #ICLR2025 attendees

Were there any updates about their LLM experiment evaluation during ICLR 2025?

I also heard that outstanding reviewers haven't been announced yet though it was promised, wondering if those surveys were conducted, or the results are hidden for some reason

Eureka Inference-Time Scaling Insights: Where We Stand and What Lies Ahead - Microsoft Research Understanding and measuring the potential of inference-time scaling for reasoning. The new Eureka study tests nine state-of-the-art models on eight diverse reasoning tasks.

All Eureka inference-time scaling insights are now available here: www.microsoft.com/en-us/resear... It was fun sharing these and more together with Vidhisha Balachandran @vidhishab.bsky.social and Vibhav Vineet at #ICLR2025.