Grigoris Chrysos
@Grigoris_c
Assistant Professor at @UWMadison. Interested in representation learning and trustworthy ML. https://grigoris.ece.wisc.edu/
Indeed, why is code not mandatory? That could solve various problems.
The NeurIPS paper checklist corroborates the bureaucratic theory of statistics. argmin.net/p/standard-err…
🚨 Panel on "how are theoretical tools useful in vision?" with an amazing list of panelists: @CevherLIONS @orussakovsky @vidal_rene Open to your questions, the more ambitious the better. In @CVPR : Room 107 A at 12 🎸.
Sparsity but with increased expertise across tens of thousands of combinations? Check out the details from James ⬇️
Sparse MLPs/dictionaries learn interpretable features in LLMs, yet provide poor layer reconstruction. Mixture of Decoders (MxDs) expand dense layers into sparsely activating sublayers instead, for a more faithful decomposition! 📝 arxiv.org/abs/2505.21364 [1/7]
🇬🇷 Greeks in AI is gaining traction! ✅ 200+ sign-ups 📄 30+ OpenReview submissions 🤝 Growing sponsor list 🚨 Limited space @ Serafeio — register now: greeksin.ai Stay tuned for updates on program, speakers & more! #GreeksInAI #AI #ML #Research #Greece #OpenReview
Thank you very much for the talk and the diverse applications in real systems.
Thanks to @Grigoris_c, @SharonYixuanLi & the other organizers for inviting me! It was fun talking about AI for Robust Decision Making at #ICLR2025. Slides here: yisongyue.com/talks/robust_d… The main technical portion is on this paper by @yeh_im_excited: arxiv.org/abs/2409.20534
Our workshop on uncertainty and hallucination is happening now in Topaz Concourse. We are kicking off with @yisongyue’s keynote on AI for Robust Decision Making.
🤔 Calling all ML Researchers and LoRA Enthusiasts! 🤔 We’re exploring some fascinating questions about LoRA (Low-Rank Adaptation) models, and we’d love your input! 🧠🚀 1️⃣ Training LoRA Models on Academic Clusters Imagine you’re working with a small academic cluster (16/24GB…
Eager to learn and obtain new insights during the workshop
Simons Institute Workshop: "Future of LLMs and Transformers": 21 talks Monday - Friday next week. simons.berkeley.edu/workshops/futu…
Our paper notifications are out! Congratulations to the authors and look forward to an exciting lineup of discussions. Stay tuned for more details! #ICLR2025
We're organzing the "Quantify Uncertainty and Hallucination in Foundation Models" workshop at #ICLR2025! 📢 Call for Papers: Submit your work by February 2, 2025 (AOE). 🔗 More details: …certainty-foundation-models.github.io Look forward to seeing your submission and participation in the…
Congratulations again!
Our paper "Low-rank finetuning for LLMs is inherently unfair" won a 𝐛𝐞𝐬𝐭 𝐩𝐚𝐩𝐞𝐫 𝐚𝐰𝐚𝐫𝐝 at the @RealAAAI colorai workshop! #AAAI2025 Congratulations to amazing co-authors @nandofioretto @WatIsDas @CuongTr95450563 and M. Romanelli 🥳🥳🥳
New open source reasoning model! Huginn-3.5B reasons implicitly in latent space 🧠 Unlike O1 and R1, latent reasoning doesn’t need special chain-of-thought training data, and doesn't produce extra CoT tokens at test time. We trained on 800B tokens 👇
Some new work from our group that I'm very excited about! What makes weak-to-strong generalization possible? We think it's all about data!
What enables a strong model to surpass its weaker teacher? 🚀 Excited to share our ICLR 2025 paper: "Weak-to-Strong Generalization Through the Data-Centric Lens"! 🧵
Can we ensure that the text classification will remain the same under (a budget for) modifications? Tl-dr: We can, but only for certain classifiers. A lot of work to be done in the domain. Check the excellent posts by Elias for our #ICLR2025 paper: 👇
Now that the ICML rush has come to an end, I would like to share that our paper "Certified Robustness Under Bounded Levenshtein Distance" has been accepted in ICLR 2025 🥳. Stay with me for a short summary of our contributions:
Looking forward to your submissions!
We have extended the submission deadline to Feb 5, 2025 (AOE). Look forward to seeing your papers!
Hey Microsoft please stop trying to invent God and just make Outlook search work.
We're organzing the "Quantify Uncertainty and Hallucination in Foundation Models" workshop at #ICLR2025! 📢 Call for Papers: Submit your work by February 2, 2025 (AOE). 🔗 More details: …certainty-foundation-models.github.io Look forward to seeing your submission and participation in the…
The workshop submission portal is now live: openreview.net/group?id=ICLR.… Looking forward to all the exciting works on hallucinations/uncertainty/out-of-distribution performance on large models. with @SharonYixuanLi @ml_angelopoulos @stats_stephen @barbara_plank @EmtiyazKhan
Can we trust LLMs that confidently generate text but overlook their own hallucinations/limitations? Let's find out in our new workshop in @iclr_conf : 🚨 ICLR workshop on Uncertainty and Hallucination on LLMs ➡️ …certainty-foundation-models.github.io 1/2
Glad to see tensors and tensor factorizations being more broadly used for architecture design. Great work!
We're the architects now. 🏗️📐.
10 challenges for frontier models. However, what is "Oscar-caliber screenplay"? Similarly, why are 10000 lines the limit? Why not 1k?
𝗔 𝗯𝗲𝘁 𝗼𝗻 𝘄𝗵𝗲𝗿𝗲 𝘄𝗶𝗹𝗹 𝗔𝗜 𝗯𝗲 𝗮𝘁 𝘁𝗵𝗲 𝗲𝗻𝗱 𝗼𝗳 𝟮𝟬𝟮𝟳: @Miles_Brundage, formerly of OpenAI, bravely takes a version of the bet I offered @Elonmusk! Proceeds to charity. Can AI do 8 of these 10 by the end of 2027? 1. Watch a previously unseen mainstream…
Join us in learning more on fine-tuning and efficiency throughout the day.
Today 8:50am at East Exhibit Hall A! Invited speakers: @Azaliamirh @jasondeanlee @tydsh @QuanquanGu Panel discussion: @danqi_chen @tri_dao @annadgoldie @QuanquanGu @btreetaiji @tydsh @leenaCvankadara