Ingvar Max Ziemann

In November 2022, I obtained my PhD at the Division of Decision and Control Systems under the supervision of Henrik Sandberg. In March, I will join UPenn as a postdoc to be hosted by Nikolai Matni and George Pappas. I am fortunate to be generously funded by a VR grant.

My work is centered on using statistical and information theoretic tools to study learning-enabled control methods. Currently, I am interested in studying how learning algorithms generalize in the context of dynamical systems. During the first half of my PhD I have worked toward characterizing fundamental limits of data-driven control problems. However, my interests are pretty broad, and I'm more than happy to chat about just about anything in machine learning, probability and applied mathematics.

Background:Before starting my PhD I obtained two sets of Master's and Bachelor's degrees in Mathematics (SU/KTH) and in Economics and Finance (SSE). During my studies, I also spent two semesters abroad, one at ENS de Lyon (Mathematics) and one in Aix-en-Provence (French Language). My Master's thesis on Model Reduction of Semistable Infinite-Dimensional Control Systems was awarded a prize (joint top 1-3) by the Stockholm Mathematics Center.

Preprints:

• Learning with little mixing (NeurIPS'22, with Stephen Tu)

• Statistical Learning Theory for Control: A Finite Sample Perspective (Anastasios Tsiamis, Ingvar Ziemann, Nikolai Matni, George Pappas)

• How are policy gradient methods affected by the limits of control? (CDC'22,best student paper award, with Anastasios Tsiamis, Henrik Sandberg and Nikolai Matni)

• Single Trajectory Nonparametric Learning of Nonlinear Dynamics (COLT'22, with Henrik Sandberg and Nikolai Matni)

• Learning to Control Linear Systems can be Hard (COLT'22, Anastatios Tsiamis, Ingvar Ziemann, Manfred Morari, Nikolai Matni, George Pappas)

• Regret Lower Bounds for Learning Linear Quadratic Gaussian Systems (with Henrik Sandberg)

• Noninvasively improving the orbit-response matrix while continuously correcting the orbit (Physical Review Accelators and Beams, with Volker Ziemann)

• On Uninformative Optimal Policies in Adaptive LQR with Unknown B-matrix (L4DC'21, with Henrik Sandberg)

• Resource Constrained Sensor Attacks by Minimizing Fisher Information (ACC'21, with Henrik Sandberg)

• On a Phase Transition of Regret in Linear Quadratic Control: The Memoryless Case (IEEE Control Systems Letters, 2020, and to be presented at CDC'20, with Henrik Sandberg)

• Regret Lower Bounds for Unbiased Adaptive Control of Linear Quadratic Regulators (IEEE Control Systems Letters, 2020, with Henrik Sandberg)
• Parameter Privacy versus Control Performance: Fisher Information Regularized Control (ACC'20, with Henrik Sandberg).
• Model Reduction of Semistable Distributed Parameter Systems (ECC'19, with Yishao Zhou).

Courses taken: Reinforcement Learning (EL2805), Differential Geometry (SF2722), Data Driven Modelling (FEL3201), Information Theory (FEO3210), Theoretical Foundations of Machine Learning (FJL 3380), High Dimensional Probability and Algorithms (Summer School), Advanced Probability (FSF3945), Information Theory for Statistics and Learning (FEO3350)