렙식

딥러닝 연구원

[fly51fly @fly51fly]

[LG] GradInf: Gradient Estimation as Probabilistic Inference
G Arya, M Huot, M Schauer, A K. Lew… [CMU & MIT & Chalmers University of Technology] (2026)
https://t.co/M9Xs5xDWUE

[Rosinality @rosinality]

https://t.co/kCtplp8EL7

Using an interpolated target between student and teacher to reduce the gap for OPD.

[Pinta AI @Pinta_AI]

We are honored to feature on EO planet, where our founder, Dion Kim, shares our journey launching into the U.S. AI-security market.

This feature marks a milestone in our pursuit to advance AI agent security across the globe.

https://t.co/thfLqF85K4

[Didier 'Dirac's ghost' Gaulin @DiracGhost]

You need to check out 'Quantum Field Theory and Differential Geometry' by Chen, a really nice history and primer on the physical idea behind topological Yang Mills theory.

Short (22 pages), clearly written and very well suited for beginners and self learners. The reference list is quite good as well, for those willing to go further.

🔗👇

[Daniel Rosiak @DanRosiak]

"Solve all open problems and finish all open research from this. Make no mistakes."

[Frank Nielsen @FrnkNlsn]

"Analysis and Geometry of Markov Diffusion Operators"

[Math Cafe @Riazi_Cafe_en]

MIT's "High-Dimensional Statistics" lecture notes

PDF: https://t.co/bDjaU97Qee

[Francesco Bertolotti @f14bertolotti]

In this work, the authors treat the training as a chaotic dynamical system, derive risk bounds wrt where weights can go if trained for a long time. Use the theory to define a generalization metric and test it on grokking. Dense but very interesting!

🔗https://t.co/AO3ABvmwHx

[Simon Coste ꙮ @__SimonCoste__]

The main idea behind generative diffusions and flow matchings (ie, reverting the heat equation to transport something onto a gaussian) had been studied by mathematicians since (at least) 2010, under the name of « Kim-Milman map », see
https://t.co/LdCahThckc
1/3

[千寻｜AI 分享 🌸 @Crypto_QianXun]

有人刚刚开源了整个常春藤盟校的数学和科学课程。

斯坦福、麻省理工、哈佛……全部课程。集中在一个地方。完全免费。

Github：https://t.co/diVa1kV3GP

[Vaishnavi @_vmlops]

Lecture Notes on Statistics and Information Theory

https://t.co/CwnrO85kia

[Stat.ML Papers @StatMLPapers]

Flow Matching is Adaptive to Manifold Structures https://t.co/L5EKv1Ujcv

[Stat.ML Papers @StatMLPapers]

Generative Path-Law Jump-Diffusion: Sequential MMD-Gradient Flows and Generalisation Bounds in Marcus-Signature RKHS https://t.co/NuzBp8xysp