Takeshi Koshizuka (越塚 毅)

SPDR @ RIKEN AIP | PhD | Physics × ML

[Takeshi Koshizuka (越塚 毅) @KoshStorm]

I’m happy to share that I’ve completed my PhD 🎓
I’m sincerely grateful to Prof. Issei Sato and everyone who supported me along the way.

I have started a position as a Special Postdoctoral Researcher (基礎特研) at RIKEN, working at the intersection of AI and science.

[Anshul Kundaje @anshulkundaje]

But the Bitter Lesson from AlphaFold & EVO2 is that scaling is actually very challenging in biology & even today models that incorporate domain specific constraints & inductive biases are still very hard to beat. 8/

[Shohei Hido 比戸将平 @sla]

2026年3月8日に東大安田講堂で開かれる国際シンポジウム「知の協奏：AI×物理」にダイキンが後援で入っています。
https://t.co/JYmtg2U0Nk

甘利先生、松尾豊先生、PFN岡野原さん、Max Welling博士など著名な招待講演者を迎えての貴重な機会です。無料イベントですのでぜひ参加ご検討ください。

[Takeshi Koshizuka (越塚 毅) @KoshStorm]

Excited to present our latest work at #NeurIPS2025 in San Diego! 🇺🇸
Check out our poster and video: https://t.co/zNvACiPnac

[Takeshi Koshizuka (越塚 毅) @KoshStorm]

🎉 Thrilled to announce our paper, "Understanding Generalization in Physics Informed Models through Affine Variety Dimensions," has been accepted to #NeurIPS2025!

paper link: https://t.co/f17PuhXOPR

A huge thanks to my supervisor, Issei Sato(@issei_sato)

[Dawei Li✈️ICLR2026 @Dawei_Li_ASU]

A bug in #OpenReview just exposed every reviewer’s identity across all conferences.😂

⚔️Next week’s in-person NeurIPS be like: 👇

[パリ好き🇫🇷 @parisfrfr]

1ユーロが179円を超えたことをお知らせします

[Schmidt Sciences @schmidtsciences]

We're excited to welcome 28 new AI2050 Fellows! This 4th cohort of researchers are pursuing projects that include building AI scientists, designing trustworthy models, and improving biological and medical research, among other areas. https://t.co/8oY7xdhxvF

[Lorenz Richter @lorenz_richter]

Solving control problems can be hard. This is why we introduce trust region methods, approaching them iteratively in a systematic way. In fact, this can be understood as a geometric annealing from prior to target with adaptive steps. More at @NeurIPSConf, https://t.co/YS1VkDYBEB.

[Google DeepMind @GoogleDeepMind]

We’re announcing a major advance in the study of fluid dynamics with AI 💧 in a joint paper with researchers from @BrownUniversity, @nyuniversity and @Stanford.

[Takeshi Koshizuka (越塚 毅) @KoshStorm]

Physics-Informed Linear Regression (PILR) の汎化性能が、モデルのパラメータ数ではなく微分方程式制約から生じるアフィン代数多様体の次元に依存することを示し、これにより「物理法則」が過学習を防ぐ理由を説明しています。

[Takeshi Koshizuka (越塚 毅) @KoshStorm]

🎉 Thrilled to announce our paper, "Understanding Generalization in Physics Informed Models through Affine Variety Dimensions," has been accepted to #NeurIPS2025!

paper link: https://t.co/f17PuhXOPR

A huge thanks to my supervisor, Issei Sato(@issei_sato)

[Takeshi Koshizuka (越塚 毅) @KoshStorm]

We show the generalization of Physics-Informed Linear Regression depends on the affine variety dimension from the physical constraints, rather than its parameter count, explaining why 'physics' can prevent overfitting.

[Kirill Neklyudov @k_neklyudov]

1/ Where do Probabilistic Models, Sampling, Deep Learning, and Natural Sciences meet? 🤔 The workshop we’re organizing at #NeurIPS2025!

📢 FPI@NeurIPS 2025: Frontiers in Probabilistic Inference – Learning meets Sampling

Learn more and submit → https://t.co/gDt9lFzPBv

Organizers: @tara_aksa @msalbergo @bose_joey @marylougab @theh2o64 @guanhorng_liu Grant Rotskoff @EvaSmorodina @AlexanderTong7

[Takeshi Koshizuka (越塚 毅) @KoshStorm]

Currently visiting ZIB in Berlin

[Kevin Xu @kevin671xu]

🎉 Our paper on "the expressive power of Looped Transformers" was accepted at #ICML2025 !
To the best of our knowledge, this is the first study to analyze their function approximation capabilities, including approximation rates and universality.
https://t.co/i90Z7KujbI

[enjoy my life @issei_sato]

The following paper has been accepted to ICML 2025!
On Expressive Power of Looped Transformers: Theoretical Analysis and Enhancement via Timestep Encoding
Kevin Xu, Issei Sato
https://t.co/uYfe1vaApm

[enjoy my life @issei_sato]

The following paper has been accepted to ICML 2025!
Benign Overfitting in Token Selection of Attention Mechanism
Keitaro Sakamoto, Issei Sato
https://t.co/nqdKAZIQD8

[enjoy my life @issei_sato]

以下の論文がICLR2025に採択されました！

[enjoy my life @issei_sato]

以下の論文がICLR2025に採択されました！長谷川さん、M2で2年連続ICLRに採択されててすごい。

[Takeshi Koshizuka (越塚 毅) @KoshStorm]

Our paper on Neural Operator: 'Understanding the Expressivity and Trainability of Fourier Neural Operator: A Mean-Field Perspective' was accepted at #NeurIPS2024 🎉 🇨🇦
paper link: https://t.co/hBB0JnM0dn

Huge thanks to @fujisawa0211 , @yusuk_et , @issei_sato