Hi, I am Dian Jin (金典), a Ph.D. student in Electrical and Computer Engineering (minor in Computer Science) at UW–Madison, advised by Prof. Jeremy Coulson. I work on data-driven prediction control: building predictive models of system behavior directly from input–output data, grounded in behavioral systems theory and DeePC. Previously, I received my B.S. (2022) from Soochow University and M.A. (2023) from UW–Madison, both in Mathematics. Outside research, I enjoy photography and the outdoors.
News
- Jul 2026 Paper “On the sensitivity of the subspace predictor to behavioral perturbations” accepted to IEEE Control Systems Letters (L-CSS), with an oral presentation in the invited session Safe Planning and Control with Uncertainty Quantification at the 65th IEEE Conference on Decision and Control (CDC 2026), Honolulu, Hawaii.
- Jan 2026 Paper “Online subspace learning on flag manifolds for system identification” accepted to the 8th Annual Learning for Dynamics & Control Conference (L4DC 2026), Los Angeles.
Research Interests
I develop data-driven methods for learning predictive models of dynamical systems from trajectory data, and using them for prediction and control:
- Behavioral clusters for predictive control — approximating trajectories of nonlinear systems with collections of low-dimensional linear subspaces, combined with graph-based reference selection and constrained predictive control. Evaluated in numerical simulation on standard benchmark tasks.
- Online subspace learning — updating low-dimensional predictive models of time-varying systems as new data arrive, via optimization on Grassmann and flag manifolds, without knowing the system order in advance.
- Sensitivity of data-driven predictors — deriving prediction-error bounds that quantify how errors in learned subspaces propagate to downstream predictions and control.
Recent Publications
IEEE L-CSS · CDC 2026
On the Sensitivity of the Subspace Predictor to Behavioral Perturbations
Oral presentation in the invited session Safe Planning and Control with Uncertainty Quantification at the 65th IEEE Conference on Decision and Control (CDC 2026).
L4DC 2026
Online Subspace Learning on Flag Manifolds for System Identification
Proceedings of the 8th Annual Learning for Dynamics and Control Conference, PMLR 331:2138–2150.
NeurIPS 2023
Parameter-Efficient Tuning of Large-Scale Multimodal Foundation Model
Advances in Neural Information Processing Systems 36, 15752–15774.
Academic Services
- Journal reviewer: IEEE Transactions on Control Systems Technology (TCST), Automatica
- Conference reviewer: L4DC’25, L4DC’26, CDC’26