This is a conjecture stated in Cyril Letrouit's Peccot course lecture (not acknowledged by OpenAI). The paper by Cyril Letrouit and Tudor Manole gives an alternative and more general proof. arxiv.org/abs/2610.37659
Gabriel Peyré
Researcher with public evidence across AI research.
- AI signals
- 7 past 30d
- Sources
- 3 distinct domains
- Discusiones
- 0 past 30d
- Latest signal
- 4h ago
Articles & links
For what is worth, this question of Wasserstein-L2 comparison is exactly one of the problems that was cracked by OpenAI: github.com/openai/math/...
And you can consider a bistochastic kernel - Sinkhorn: arxiv.org/abs/2603.12366 (which is a sinkformer arxiv.org/abs/2110.11773) which leads to an almost perfect translation of the mean.
And you can consider a bistochastic kernel - Sinkhorn: arxiv.org/abs/2603.12366 (which is a sinkformer arxiv.org/abs/2110.11773) which leads to an almost perfect translation of the mean.
It is a comparison between - MMD: a kernel (2017) arxiv.org/abs/1801.01401 - Drifting: a stochastic kernel (2026) arxiv.org/abs/2602.04770
It is a comparison between - MMD: a kernel (2017) arxiv.org/abs/1801.01401 - Drifting: a stochastic kernel (2026) arxiv.org/abs/2602.04770
For least squares with constant noise variance, fixed-step SGD has an isotropic stationary covariance to first order. Fourth moments determine the next correction. https://www.gpeyre.com/blog/2026/11/18/sgd-iterate-covariance/
While studying the convergence of linear neural networks, I stumbled upon a nice conservation law that depends only on the distribution of neurons: a “mean-field” conservation law. www.gpeyre.com/blog/2026/10...
Slides for an introductory talk on IA for maths. speakerdeck.com/gpeyre/ia-fo...
How do max and softmax layers approximate convex functions? Supporting planes give a constructive proof, error bounds, and a tradeoff between neuron count and temperature. https://www.gpeyre.com/blog/2026/10/24/convex-max-softmax-approximation/
Approximating convex functions is hard; input-convex neural networks (ReLu networks with positive weights) are convex, but for finite depth, they are not universal approximators. www.gpeyre.com/blog/2026/10...
Un court article grand public pour CNRS le Mag : Le Jour où… « L’IA a fait des maths mieux que moi » www.gpeyre.com/blog/2026/11... (with also an english translation)
Recent commentary
Thanks to LLMs, I think the exponent for the (refutal) of Erdos unit distance conjecture will be more famous than omega (the exponent for matrix multiplication)!
In Gabriel Peyré's orbit
Center = Gabriel Peyré. Left = members they follow (green edges). Right = members who follow them (blue edges). Top = mutual follows (orange edges, slightly larger). Drag any node to reposition; click to open that profile.
Are you Gabriel Peyré? Show it.
Add the Who’s Who of AI badge to your site or bio. It links back to this profile.
Markdown: [](https://aiweekly.co/whos-who/person/gabrielpeyre-bsky-social)