Charles Riou
@charles-riou.bsky.social
I share my love for machine learning on my YouTube channel ML New Papers: https://youtube.com/@mlnewpapers-xo5vm?si=3Y1SFI2yV_o8i89n Follow if you like ML! PhD in CS/ML (University of Tokyo) Eng. Diploma from École Polytechnique (France)
Heart Aerospace just flew the world’s largest battery-electric aircraft! 🔹27 min flight 🔹12 tons 🔹$5 electricity 🔹Up to 200 km range Electric long-haul flight is still limited by battery energy density. But could hybrid-electric aircraft do with the skies what electric cars did with the roads? ✈️
August 14, 2026 marked French cartoonist/writer René Goscinny’s 100th bday. His colorful phrases, like in the Astérix comics, inspired my own YouTube scripts (yes, I wrote them myself - ils sont fous, ces YouTubers). Happy bday to him & happy Assumption Day if you celebrate it!
Hi all! I just released a YouTube vlog to AISTATS 2025, one of the major AI conferences. And this year, it is in Thailand 🌴 Machine learning and elephants aficionados alike, link to the video in the comments ⬇️
Researchers from CNRS & ESSEC in France will travel to their respective Singapore campuses in June. We are taking the opportunity to organize a mini-workshop on stats/ML at ESSEC Singapore on Friday, June 13. If you wish to attend, you are welcome, but for logistic reasons please let me know 🙏
We know the shapes of all human proteins! That’s a definitive step in healthcare… but why? I explain that in my latest YouTube video on the Nobel Prize in Chemistry 2024, awarded to Baker, @demishassabis and @JohnJumperSci from @GoogleDeepMind ⬇️ link the comments
I just did a Y0uTube interview of @pierrealquier.bsky.social, Professor of Statistics, and expert in statistical machine learning theory. What is the state of current AI theory? What are his tips to present his work efficiently and convince his audience in AI? Link in the comments ⬇️
Little history of science fact: exactly 70 years ago, on April 18th 1955, Albert Einstein passed away at age 76. He did research until his last day, aiming to develop a theory unifying the laws of physics.
AI takes the Nobel Prize cake: math, physics, chemistry… and now computer science, with the Turing Award going to RL founders @RichardSSutton and Andrew Barto! Wanna know more? Check out my Y0uTube video (link in the comments ⬇️)
How can AI and modern LLMs like ChatGPT generate a chain of thoughts to solve difficult problems? A recent work by Juno Kim and Taiji Suzuki is a significant step toward understanding this, which I explain in a Y0uTube video (link in the comments ⬇️)
Do you procrastinate? As a PhD student, I used to have 7 nasty habits that ruined my productivity for months. Wanna know what those are and how I became productive again? Check out my Y0uTube video in the comments ⬇️ , and subscribe if you are interested in AI.
The meta learning always achieves a fast learning rate (at the meta level)! And how washing the dishes yesterday will help me to brush my teeth more efficiently tomorrow… Wanna know more? Check out our joint work with Pierre Alquier @pierrealquier.bsky.social and Badr-Eddine Chérief-Abdellatif.
Pierre Alquier@pierrealquier.bsky.social · 2y agoPublished today (joint work with Charles Riou @charles-riou.bsky.social and Badr-Eddine Chérief-Abdellatif). www.jmlr.org/papers/v26/2...
A simple coin-betting game can help optimize neural nets. More generally, it can help derive a bound on the error of a Bayes posterior, i.e. a PAC-Bayes bound, which is better and involves a new divergence. Wanna know how, check out the link in the comments ⬇️
PhD student here, working on my research at Starbucks. What happened to the music playlist today? It would depress a rabbit in a carrot farm.
Our paper on Bayesian meta-learning with Charles Riou @charles-riou.bsky.social (University of Tokyo) and Badr-Eddine Chériéf-Abdellatif (@cnrs.bsky.social) accepted today for publication in the Journal of Machine Learning Research 😊 arxiv.org/abs/2302.11709
Bayes meets Bernstein at the Meta Level: an Analysis of Fast Rates in Meta-Learning with PAC-Bayes
Bernstein's condition is a key assumption that guarantees fast rates in machine learning. For example, the Gibbs algorithm with prior $π$ has an excess risk in $O(d_π/n)$, as opposed to the standard $...
arxiv.org
The 2025 Calendar & Map of machine learning conferences is OUT! Wanna know where and when the ML community will gather this year? Feel free to check my Y0uTube video and its description for a pdf (link in the comments).
If like me, you looove math, machine learning, statistics or computer science, follow me and I will follow you back!