Dynamai

Dynamai is a fundamental machine learning (funML) research lab built on applied and computational mathematics. We focus on problems at the interface of applied probability, dynamical systems, and ML. Our broader vision is to shape the next generation of ML by advancing mathematical foundations and principled modeling approaches. Our research spans generative modeling, sequence modeling, robustness and reliability in ML, ML for science, and beyond.

Discover and explore our latest research on arXiv and our open-source projects on GitHub.

Selected News

📣 Ongoing: Our lab is expanding! Several openings for highly motivated master’s/PhD students and postdocs.
📣 February 2026: Two preprints were posted: Is Flow Matching Just Trajectory Replay for Sequential Data? and A Kinetic-Energy Perspective of Flow Matching.
📣 December 2025: One preprint on On The Hidden Biases of Flow Matching Samplers was posted.
📣 July 2025: Our paper on Elucidating the Design Choice of Probability Paths in Flow Matching for Forecasting was accepted at TMLR.
📣 May 2025: One preprint on FLEX: A Backbone for Diffusion-Based Modeling of Spatio-temporal Physical Systems was posted.
📣 April 2025: Congratulations to Shizheng Lin on being selected for the Digital Futures Summer Research Internship Programme 2025!
📣 Jan 2025: Our paper on Tuning Frequency Bias of State Space Models was accepted (spotlight) at ICLR 2025.
📣 Jan 2025: Our paper on Gated Recurrent Neural Networks with Weighted Time-Delay Feedback was accepted at AISTATS 2025.

Funding & Support

We acknowledge the computational resources provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS), partially funded by the Swedish Research Council through grant agreement no. 2022-06725. We are very grateful for the funding awarded by the Swedish Research Council (VR/2021-03648) and the Wallenberg Foundations (WINQ).

Contact

Interested in collaborating with and/or supporting our research? Please get in touch at dynamai DOT lab AT gmail.com.