I develop and implement algorithms to simulate, learn and control dynamical systems: inferring their states and parameters from noisy, partial observations, and choosing actions that steer them toward a goal. With Dynestyx, our open-source probabilistic programming library, a model is written once and can then be simulated, filtered, smoothed, calibrated or controlled, across discrete- and continuous-time, deterministic and stochastic systems.
Dynestyx
I am a developer of Dynestyx, an open source library for simulation, inference and control in dynamical systems. Here are some problems we can solve in Dynestyx:
State inference and system ID in robotics. A robot hand grasps an object of unknown friction and mass in MuJoCo; we infer both from the observed trajectory. Data assimilation for complex 3D PDEs. Filtering a turbulent 3D Navier–Stokes flow from coarse observations: true state, observation and filtered estimate. Simulating complex environments coupled with control policies. A swarm of particles steered by a closed-loop controller in the wake of a cylinder, simulated jointly with the 2D fluid flow.
Relevant research
Dynestyx: an open source library for dynamical systems
Probabilistic programming for dynamical systems, built on NumPyro: deterministic and stochastic, discrete- and continuous-time models.
@article{darcy2023oneshot,title={One-Shot Learning of Stochastic Differential Equations with Data Adapted Kernels},author={Darcy, M. and Hamzi, B. and Livieri, G. and Owhadi, H. and Tavallali, P.},journal={Physica D: Nonlinear Phenomena},volume={444},pages={133583},year={2023},doi={10.1016/j.physd.2022.133583},}
Learning Dynamical Systems from Data: A Simple Cross-Validation Perspective, Part II: Nonparametric Kernel Flows
M. Darcy, B. Hamzi, J. Susiluoto, and 2 more authors
@article{darcy2025kernelflows,title={Learning Dynamical Systems from Data: A Simple Cross-Validation Perspective, Part {II}: Nonparametric Kernel Flows},author={Darcy, M. and Hamzi, B. and Susiluoto, J. and Braverman, A. and Owhadi, H.},journal={Physica D: Nonlinear Phenomena},volume={476},pages={134641},year={2025},}
Benchmarking Gaussian Processes for Prediction and Data Assimilation of Alzheimer’s Disease Progression
P. Batlle, M. Darcy, M. Levine, and 6 more authors
@article{batlle2025alzheimer,title={Benchmarking {G}aussian Processes for Prediction and Data Assimilation of {A}lzheimer's Disease Progression},author={Batlle, P. and Darcy, M. and Levine, M. and Hamzi, B. and Owhadi, H. and Bastani, M. and Zhang, Y. and Sarsour, K. and Talamas, F.},journal={Journal of Alzheimer's Disease},volume={109},pages={1137--1141},year={2025},doi={10.1177/13872877251404082},}
