Mathieu Blondel
@mblondel_ml
Followers
9K
Following
2K
Media
116
Statuses
4K
Research scientist at Google DeepMind. Current research interests: differentiable programming, LLMs, Transformers.
Paris, France
Joined June 2009
We uploaded V3 of our draft book "The Elements of Differentiable Programming". Lots of typo fixes, clarity improvements, new figures and a new section on Transformers!
7
75
447
Building stuff from scratch is the best way to learn!.
Wrapped up Stanford CS336 (Language Models from Scratch), taught with an amazing team @tatsu_hashimoto @marcelroed @neilbband @rckpudi. Researchers are becoming detached from the technical details of how LMs work. In CS336, we try to fix that by having students build everything:.
0
0
3
Back from MLSS Senegal 🇸🇳, where I had the honor of giving lectures on differentiable programming. Really grateful for all the amazing people I got to meet 🙏 My slides are here
2
17
63
RT @nileshtrivedi: @mblondel_ml I gave a talk based on this book. Thanks for the new version. 🙏 . Here are the slides: .
0
1
0
Can a book ever become typo-free? T_T Thank you so much to everybody who caught typos and reported them!.
0
0
0
RT @nileshtrivedi: Finishing my slides for a talk on the Elements of Differentiable Programming this Wednesday at @lossfunk . Most material….
0
28
0
Slides of my talk on our ICML 2025 paper "Joint Learning of Energy-based Models and their Partition Function"
1
14
118
Here is a table that helps to situate our approach in relation to existing approaches.
0
1
5
Application to dynamic vehicle routing. First paper of Germain's PhD (co-advised by @ParmentierAxel1 and I): really nice work at the interface of statistics and optimization!.
1
0
1
We just released a new approach for turning local search heuristics used to solve NP-hard combinatorial problems in OR into differentiable layers. The key idea is to use the neighborhoods used by these algorithms for creating MCMC proposal distributions
2
26
190
RT @s_scardapane: For those asking for material - I will mostly be following the amazing "elements of differentiable programming" ( https://….
0
18
0
RT @afra_amini: Current KL estimation practices in RLHF can generate high variance and even negative values! We propose a provably better e….
0
29
0
RT @leanprover: Fascinating talk by Thomas Hubert on AlphaProof at IMO 2024! Combining Lean's formal verification with DeepMind's RL techni….
0
41
0
RT @ramealexandre: Hiring two student researchers for Gemma post-training team at @GoogleDeepMind Paris! First topic is about diversity in….
0
36
0
I like the view in @karpathy 's latest video of LLM chat bots as using a discrete symbol communication channel: the user and the model (and also internet search) take turns inserting tokens in the channel and the model uses this context for generating its next responses.
0
0
9