Yoshinobu Kawahara

Professor, Graduate School of Information Science and Technology, The University of Osaka (@UOsaka_en) / Team Director, RIKEN AIP Center (@RIKEN_AIP_EN)

[Yoshinobu Kawahara @44nb_k]

Our paper "Mesh Field Theory: Port-Hamiltonian Formulation of Mesh-Based Physics" (first author: S. Noguchi (JAMSTEC / RIKEN AIP)) was accepted to @icmlconf #ICML2026 (arXiv:2605.00394 https://t.co/ai6Ij9m12d). For ML-based prediction of physical dynamics on meshes, Mesh Field Theory fixes the topological skeleton derived from universal physical principles and learns only the metric and dissipative structures from data, enabling structure-preserving, physically faithful, and data-efficient simulation.

[Yoshinobu Kawahara @44nb_k]

さきがけ「デジタル時空間拡張基盤」領域が立ち上がりました。数理科学・情報科学・計算科学・データ科学の広い分野が関連すると思いますので、領域ページhttps://t.co/eHCjZPFaJDをご確認の上で、ぜひ応募ご検討ください。

[JST CREST・さきがけ・ACT-X @JST_Kisokenkyu]

2026年度発足のCREST・さきがけ・ACT-X 新規研究領域はこちらです。

各領域の詳細は、募集HPの「提案を募集する研究領域」からそれぞれの領域のページに入ってご確認ください。
https://t.co/Dla62Yckb0

[Yoshinobu Kawahara @44nb_k]

@Nonline21515552 クープマン作用素の連続スペクトルの影響を利用したティッピング予測 (検知) の方法に関する論文です。研究室M2の宮内優太くんが第一著者の論文です。

[Yoshinobu Kawahara @44nb_k]

Our new paper titled "Generalized Stochastic Resilience for Early Warning Signals Based on Koopman Operator" is now out in Nonlinear Dynamics @Nonline21515552, #NonlinearDynamics (first author: Yuta Miyauchi from The University of Osaka): https://t.co/wybk81SMwn We generalize stochastic resilience for early warning signals (EWS) with stochastic Koopman operator and propose ResKMD (stochastic residual of Koopman mode decomposition), which is computable via ResDMD (residual dynamic mode decomposition), to separate noise-driven fluctuations from continuous-spectrum effects near bifurcations, enabling robust tipping-point prediction even with observation noise.

[人工知能学会 jsai-ann ML @jsai_ml]

2025年度AIPシンポジウム成果発表会のご案内
　https://t.co/HV6zl0Kzzd

[Yoshinobu Kawahara @44nb_k]

@chenzch Thanks! Yes, it’s actually closely related to Fredformer. I’ll also let Bai-san know that Chen-san sent this message, so if we have the chance, let’s discuss it.

[Yoshinobu Kawahara @44nb_k]

Our paper "Dualformer: Time-Frequency Dual Domain Learning for Long-term Time Series Forecasting" (first author: Jingjing Bai (UOsaka)) has been accepted at #AISTATS2026 (arXiv: https://t.co/yQ1FhVvTKR). We tackle Transformers' low-pass bias in long-term time series forecasting via a time/frequency dual-branch model with layer-wise band allocation.

[Yoshinobu Kawahara @44nb_k]

研究室D2の白晶晶さんが第一著者の論文です。

[Yoshinobu Kawahara @44nb_k]

@koh_t こちらこそ引き続きよろしくお願いします！

[Yoshinobu Kawahara @44nb_k]

結構マニアックですが、前から気になっていたラフパス理論を、確率的な連続深さニューラルネット (Neural SDEs) に応用した論文です。

[Yoshinobu Kawahara @44nb_k]

Our new paper titled "Infinitely deep Bayesian neural network with signature transform" is now available online in Neurocomputing (first-authored by Dr. X. Yang from Kyushu Univ.): https://t.co/qC8KwCOmM2 Building on rough path theory, we study a continuous-depth Bayesian neural network and show that applying the signature transform to Neural SDEs can improve the stability and robustness of their stochastic properties.

[Yoshinobu Kawahara @44nb_k]

今回NeurIPS発表論文のタイトルは「Wavy Transformer」ですが、主著の野口さん (JAMSTEC/RIKEN) 曰く、wavyは俗語で、イケてる、イカした、みたいな意味があるらしい (元来の意図は波動的な)。そう考えると結構攻めたタイトルかもですね。。ポスターは5日昼です。https://t.co/Fk2GKWDMoH #NeurIPS2025

[Yoshinobu Kawahara @44nb_k]

お、ほんまや

[ぃすぁを @iSuawo]

お、めっちゃ面白そうな論文がNonlinearityから出てる👀

Koopman operators with intrinsic observables in rigged reproducing kernel Hilbert spaces https://t.co/UhU79eLQlm @ioppublishingより

[Karsten Borgwardt @kmborgwardt]

Several postdoc positions and positions for scientific software engineers will become available in my department at the Max Planck Institute of Biochemistry in Munich over coming months. Topics include machine learning in bioengineering, in proteomics and in medicine. 1/2

[大阪大学 @UOsaka_ja]

【ニュース】 大阪大学の公式英語名称変更のお知らせ －The University of Osaka へ－ https://t.co/24k3XT2Pnp #阪大 #ニュース

[大阪大学 @UOsaka_ja]

【ニュース】 工学部、基礎工学部の入学定員を令和7年度より増員！～基礎理論と実践力を兼ね備えた高度情報専門人材の育成を強化！ https://t.co/hNU6VkwF6o #阪大 #ニュース

[Amir-massoud Farahmand @SoloGen]

This paper introduces an Optimal Control/Reinforcement Learning framework that minimizes the sum of per-step costs (as usual in OC/RL) + some functions of the spectrum of the Koopman operator (new).
Why should we do that? Read on!

[理化学研究所（理研） @RIKEN_JP]

理化学研究所におけるBOOSTプログラムの活用方針について

https://t.co/231aooxr5d

#科学道

[Yoshinobu Kawahara @44nb_k]

@yasuotabei 大事なとこで笑 いずれにしても、また東京に行ったらお祝いしましょう！