UOsaka Machine Learning & Systems Laboratory -- HOME --

Our laboratory conducts research on machine learning, a foundational field of information science that underpins AI technologies rapidly being adopted across society and industry. In addition to exploring new principles, algorithms, and models in this rapidly evolving field, we pursue research that applies these advances to a wide range of scientific and engineering problems, with the aim of generating new scientific insights and solving real-world challenges. Recent representative research topics include further advances in deep learning, representation learning, and reinforcement learning; machine learning for dynamical data-generating processes; and the development of machine learning models and algorithms that leverage dynamics and physical structure. We also work on applications in medical, fluid and quantum sciences, as well as machine-learning-based surrogate modeling and design automation for industrial processes, and applications to robotics and animation.

We are recruiting students who are interested in machine learning and its related fields!

Visit for the information about Dynamical Systems Learning Team led by Prof. Kawahara in RIKEN AIP Center.

Recent News & Announcements

Our paper on oscillatory recurrent neural networks titled "Kernel Occupation Readout for Oscillatory Recurrent Neural Networks" (first-authored by Y. Inui (UOsaka)) was accepted at the Int'l Joint Conf. on Neural Networks (IJCNN'26) (March 21, 2026).
Our paper on fair representation learning titled "Predictive Fair Representation Learning with Variational Autoencoders" (first-authored by T. Yamada (UOsaka)) was accepted for publication in New Generation Computing (March 10, 2026).
Our paper on curriculum learning for neural quantum states titled "Explorative Curriculum Learning for Strongly Correlated Electron Systems" (first-authored by K. Yamazaki (UOsaka)) was accepted for publication in ACM Trans. on AI for Science (ACM TAIS) (February 22, 2026).
Our paper on Transformer for long-term time-series forecasting titled "Dualformer: Time-Frequency Dual Domain Learning for Long-term Time Series Forecasting" (first-authored by J. Bai (UOsaka)) was accepted at the 29th Int'l Conf. on Artificial Intelligence and Statistics (AISTATS'26) (January 23, 2026).
Our paper on an application of ML to satellite data analysis titled "Global land-Atmospheric Data assimilation based on Koopman operator error covariance in a deep neural network latent space for GHG emission verification" (first-authored by H. Arai (UOsaka)) was accepted as a book chapter in Agriculture, Forestry and Other Land Use (AFOLU) Emissions in South and Southeast Asia (eds.: Krishna Vadrevu, Toshimasa Ohara, and Chris Justice), Springer (January 23, 2026).
Our paper on deep learning using rough path theory titled "Infinitely deep Bayesian neural network with signature transform" (first-authored by X. Yang (Kyushu Univ.)) was accepted for publication in Neurocomputing (Dec. 28th, 2025).
Our paper on tipping prediction by Koopman operator titled "Generalized Stochastic Resilience for Early Warning Signals Based on Koopman Operator" (first-authored by Y. Miyauchi (The University of Osaka)) was accepted for publication in Nonlinear Dynamics (Nov. 22th, 2025).
Our paper on Koopman analysis of dynamical systems titled "Koopman operators with intrinsic observables in rigged reproducing Kernel Hilbert spaces" (first-authored by Dr. Isao (Kyoto Univ.)) was accepted for publication in Nonlinearity (Nov. 10th, 2025).
Prof. Kawahara was named to Stanford University’s World’s Top 2% Scientists (2024) list (October 15th, 2025).
One master student, three doctoral students and one research student have joined our lab as new members (October 1st, 2025).
Three papers have been accepted by NeurIPS'25 (September 19th, 2025).
Prof. Kawahara will serve as a Senior Area Chair at the 29th Int'l Conf. on Artificial Intelligence and Statistics (AISTATS'26).
Our paper on discovery of differential equations titled "Bayesian Model Selection for Parametric Variable-coefficient Differential Equation Discovery" (first-authored by Dr. P. Thanasutives (RIKEN)) was accepted for publication in Results in Engineering (August 25, 2025).
Our paper on the convergence of approximate inference based on Stein discrepancy titled "On the Convergence of SVGD in KL Divergence via Approximate Gradient Flow" (first-authored by Assist. Prof. Fujisawa and Lect. Futami (U. of Osaka / RIKEN)) was accepted for publication in Transactions on Machine Learning Research (August 12, 2025).
Our paper on deep learning architecture titled "Disappearance of Timestep Embedding: A Case Study on Neural ODE and Diffusion Models" (first-authored by Dr. B.J. Kim (UTokyo)) was accepted for publication in Transactions on Machine Learning Research (May 22nd, 2025).
Two papers have been accepted by 42nd Int'l Conf. on Machine Learning (ICML'25) (May 19th, 2025).
Asst. Prof. Fujisawa received the 2024 RIKEN Meihou Prize (March 17, 2025).
Dr. M. Fujisawa (Asst. Prof.) has joined our lab as a new member (March 3, 2025).
Dr. M. Ikeda (Asso. Prof.) and one administrative assistant have joined our lab as new members (February 1, 2025).
The paper on dynamical systems learning, "Learning Stochastic Nonlinear Dynamics with Embedded Latent Transfer Operators" (first-authored by N. Ke and R. Tanaka (Osaka University)), was accepted at the 28th Int'l Conf. on Artificial Intelligence and Statistics (AISTATS'25) (January 22, 2025).
The paper on reinforcement learning, "Self-supervised Color Generalization in Reinforcement Learning" (first-authored by M. Weissenbacher (RIKEN)), was accepted in Transactions on Machine Learning Research (October 28, 2024).
The paper on neural mechanisms incorporating dynamical structure, "Learning with Almost Invariant Sets in Neural Oscillatory ODEs" (first-authored by Y. Inui (Osaka University)), was accepted at the 31st Int'l Conf. on Neural Information Processing (ICONIP'24) (August 21, 2024).
The paper on extended Koopman analysis, "Enhancing Spectral Analysis in Nonlinear Dynamics: An Approach Based on Pseudoeigenfunctions from Continuous Spectra" (first-authored by Isushi Sakata (RIKEN)), was accepted in Scientific Reports (August 9, 2024).
The paper on anomaly detection applications, "Graph Deep Learning-based Anomaly Detection and Root Cause Estimation in Batteries" (first-authored by Joji Yoshikawa (Kyocera)), was accepted in the IEEJ Transactions (June 13, 2024).
The paper on online learning with Koopman spectra, "Koopman Spectrum Nonlinear Regulators and Efficient Online Learning" (first-authored by Motoya Onishi (University of Washington)), was accepted in Transactions on Machine Learning Research (May 22, 2024).
The paper on data analysis using dynamic mode decomposition, "Fast, accurate, and interpretable decoding of electrocorticographic signals using dynamic mode decomposition" (first-authored by Ryohei Fukuma (Osaka University)), was accepted in Communications Biology (May 6, 2024).
The paper on reinforcement learning, "SiT: Symmetry-invariant Transformers for Generalisation in Reinforcement Learning" (first-authored by M. Weissenbacher (RIKEN)), was accepted at the Int'l Conf. on Machine Learning (ICML'24) (May 2, 2024).
The paper on interpretable machine learning models, "MANet: Mixed Attention Network for Visual Explanation" (first-authored by Jingjing Bai (Kyushu University)), was accepted in New Generation Computing (February 22, 2024).
The paper on analyzing multi-agent behavior using machine learning, "Estimating counterfactual treatment outcomes over time in complex multi-agent scenarios" (first-authored by Keisuke Fujii (Nagoya University)), was accepted in IEEE Transactions on Neural Networks and Learning Systems (January 28, 2024).
The paper on EEG analysis using dynamic mode decomposition, "Manifold alteration between major depressive disorder and healthy control subjects using dynamic mode decomposition in resting-state fMRI data" (first-authored by Hidenori Endo (ATR)), was accepted in Frontiers in Psychiatry (January 15, 2024).
The paper on reinforcement learning-based modeling of multi-agent behavior, "Adaptive action supervision in reinforcement learning from real-world multi-agent demonstrations" (first-authored by Keisuke Fujii (Nagoya University)), was accepted at the 17th Int'l Conf. On Agents and Artificial Intelligence (ICAART'24) (December 8, 2023).
The paper on data-driven analysis of cellular automata using dynamic mode decomposition, "Dynamic mode decomposition for Koopman spectral analysis of elementary cellular automata" (first-authored by Keiri Taga (Waseda University)), was accepted in Chaos (December 1, 2023).
The paper on imitation learning for multi-agent behavior modeling, "Decentralized Policy Learning with Partial Observation and Mechanical Constraints for Multi-person Modeling Learning Systems" (first-authored by Keisuke Fujii (Nagoya University)), was accepted in Neural Networks (November 30, 2023).
Professor Shunji Umetani (Invited Professor) and Mr. Xiaoyu Yang (Special Postdoctoral Researcher) have joined our lab as new members (October 2, 2023).
The paper on tensor decomposition, "Many-body Approximation for Non-negative Tensors" (first-authored by Kazushi Garamkali (RIKEN AIP)), was accepted at the 26th Conf. on Neural Processing Systems (NeurIPS'23) (September 22, 2023).
The paper on anomaly detection, "A Characteristic Function for Shapley-Value-Based Attribution of Anomaly Scores" (first-authored by Naoya Takeishi (University of Tokyo)), was accepted in Transactions on Machine Learning Research (July 11, 2023).
The paper on statistical methods using reproducing kernels, "euMMD: Efficiently Computing the MMD Two-Sample Test Statistic for Univariate Data" (first-authored by D. Bodenham (Imperial College London)), was accepted in Statistics and Computing (June 22, 2023).
One research student has joined our lab as a new member (June 19, 2023).
The paper on deep equilibrium models (DEQ), "Stable Invariant Models with Koopman Spectra" (first-authored by Takuya Konishi), was accepted in Neural Networks (May 20, 2023).
Two administrative assistants and one research student have joined our lab as new members (May 1, 2023).
Four 4th-year undergraduate students from the Department of Information and Computer Sciences, School of Engineering Science, have joined our lab as new members (April 17, 2023).
Professor Kawahara has been appointed as a Domain Advisor for the JST ACT-X program "Innovative Mathematical and Information Science for the Next-Generation AI" (April 11, 2023).
One specially appointed lecturer, one doctoral student, and four special research students have joined our lab as new members (April 3, 2023).