←
Laurent Charlin191 posts
Laurent Charlin avatar

Laurent Charlin

@lcharlin

Machine learning professor at Mila & HEC Montréal.

◷ Joined December 2009▧ 4 Photos and videos
Laurent Charlin avatar
Laurent Charlin@lcharlin·Nov 5

This work, led by Anirudh, opens a new way to consider test-time scaling and its trade-offs and provides actionable insights.

1/n Does scaling test-time compute on LLM self-verification keep paying off? 🤔 Not really. Our work shows scaling compute makes judgements more consistent, but not necessarily more correct 🧵 Paper: arxiv.org/abs/2611.00350 Blog: anirudhb11.github.io/blog/sc… Code: github.com/anirudhb11/scorpi…

Laurent Charlin avatar
Laurent Charlin@lcharlin·Jul 6

We’re coming out of stealth with some exciting AI-optimized compute results. Some highlights: 1) Works on industrial-grade and consumer-grade GPUs. 2) It can optimize different compute layers, from the kernels to tokens, for higher tokens-per-watt on your GPU.

The entire industry is racing to make AI cheaper. Most people assume the answer is better chips or bigger GPU clusters. However, while the industry is buying its next 2× when there's a 1.78× speedup is sitting in the software. We've been hard at work @unifiedsciences, building an autonomous optimization engine toward radically more efficient compute. Today I'm very happy to share some early results where our unified engine: 1) wrote faster GEMM kernels for cuBLAS (up to 1.78x faster) 2) made an efficient AI model 6x smaller and 5x faster without any quality loss 3) optimized diverse everyday AI workloads on a consumer GPU!

Post attachment
0:28
Paul Vicol avatar
⟳ Laurent Charlin retweeted
Paul Vicol@PaulVicol·Mar 10

🎨 Introducing the TMLR Beyond PDF Editor, a web-based editor for creating Beyond PDF submissions! 🚀 Zero setup + live preview + download & import. 🤝 Compatible with local compilation. 🎉 Thanks to @hugo_larochelle @thegautamkamath @NailaMurray Nihar B. Shah @lcharlin!

0:36
Emiliano Penaloza avatar
⟳ Laurent Charlin retweeted
Emiliano Penaloza@emilianopp_·Feb 19

Problem: When your base model is basically clueless at a task, how do we get any RL signal? Solution: Use the guidance (e.g., logprobs) of a privileged teacher to guide RL. Over the past few weeks, MANY papers have converged on this idea. This comprehensive blog breaks down self-distillation and privileged distillation objectives and gives clean intuition on how they work. Link to X article, with the link to the full post in the 🧵(recommended for best experience). x.com/emilianopp_/status/202…

Post attachment

Emiliano Penaloza @emilianopp_ Feb 19 Article Understanding Self-Distillation and Privileged Information Distillation This post is an overview of methods used in recent self-distillation and privileged information distillation works. The goal is to build intuition visualizing how they work (in a simplified way). The

Emiliano Penaloza avatar
⟳ Laurent Charlin retweeted
Emiliano Penaloza@emilianopp_·Feb 6

Remember all the self-distillation papers that came out last week. Well, we also propose it 😅, but… But alongside something better 😎 π-Distill We show that with this method, you can distill closed-source frontier models even tho their traces are hidden 🔒. Both our methods can reach and even surpass the performance of the industry-standard SFT + RL with access to reasoning traces 🤯. 🔬And we spent ~100,000 hours GPU hours on a comprehensive analysis, not because the method is finicky, but because we wanted to understand why it works so well. 🧵 1/10

1:15
Yipeng Zhang avatar
⟳ Laurent Charlin retweeted
Yipeng Zhang@yipengzz·Feb 3

How can we predict multiple plausible targets from a single context in joint-embedding self-supervised learning (SSL)? Check out our paper titled “Self-Supervised Learning from Structural Invariance” accepted at #ICLR2026! Previously Best Paper Award at @unireps 2025. arxiv.org/abs/2602.02381 We introduce AdaSSL, which models the target uncertainty and relaxes the standard assumption that the positive pair share the same semantic features. Derived from first principles, we realize @ylecun’s JEPA with a learned latent variable for jointly learning better representations and world models, extending SSL’s utility to a broader range of data types. 1/🧵

Gaurav Sahu avatar
⟳ Laurent Charlin retweeted
Gaurav Sahu@dem_fier·26 Mar 2025

🚀 Exciting news! Our work LitLLM has been accepted in TMLR! LitLLM helps researchers write literature reviews by combining keyword+embedding-based search, and LLM-powered reasoning to find relevant papers and generate high-quality reviews. LitLLM.github.io/ 🧵 (1/5)

0:34
Laurent Charlin avatar
Laurent Charlin@lcharlin·5 Dec 2024

Thank you so much @CIFAR_News for your continued support through the AI Chair Program. It's an honour! It allows me to recruit fantastic students and spend more time doing research. I look forward to contributing to the success of the program. Congrats to my colleagues!

Congratulations also to @arbtal, @pierrelux, @apsarathchandar @lcharlin, @sivareddyg and @irinarish, all Canada CIFAR AI Chairs at @Mila_Quebec, on their renewal 🎉

Laurent Charlin avatar
Laurent Charlin@lcharlin·15 Nov 2024

Consider submitting your CL-related work to CoLLAs 2025 (deadline Feb: 21)!

📢 Exciting News! The Fourth Conference on Lifelong Learning Agents (CoLLAs 2025) will be held at the University of Pennsylvania (@Penn) in Philadelphia, USA 🇺🇸 🗓️ Important Dates: Abstract Deadline: Feb 21, 2025 Submission Deadline: Feb 26, 2025 Conference Dates: Aug 11 - Aug 14, 2025 We invite submissions that present new theories, methodologies, applications, or insights into algorithms and benchmarks designed for non-i.i.d. and non-stationary settings. Accepted papers will be published in the Proceedings of Machine Learning Research (PMLR). 📚 Full CFP: lifelong-ml.cc/Conferences/2… #CoLLAs2025 #AI #MachineLearning #ContinualLearning #LifelongLearning #ResearchConference #CallForPapers #NonStationaryLearning

Post attachment
Laurent Charlin avatar
Laurent Charlin@lcharlin·11 Nov 2024

+ It can be adapted to a wide range of tasks, and we apply it to frequency estimation in data streams. Curious to hear about real-world applications where these ideas could be helpful...