Learning in the Loop: Control Under Distribution Shift

▷ Call for lightning talks: Six 10-minute slots are reserved for PhD students and postdoctoral researchers. To participate, send a one-page abstract to cdc26workshop@gmail.com (plus a full-paper link if available) by October 4, 2026, 23:59 AoE. Submissions will be lightly reviewed, with decisions communicated by mid-October.

About the Workshop

Data-driven methods have become a cornerstone of modern control design, enabling remarkable progress across robotics, autonomous systems, and other safety-critical applications. Yet one fundamental challenge remains: the data available during training is rarely representative of the conditions encountered in deployment. This distributional shift can both degrade the performance of learned controllers and invalidate the guarantees used to certify their safety and reliability.

This workshop brings together leading researchers from control, optimization, and statistical learning to discuss the theoretical foundations and practical implications of control under distributional shift. Topics range from distributionally robust optimization and learning-based control to certification methods and the unique challenges posed by feedback-induced distribution shifts, where the controller itself changes the data it observes.

Through invited talks by leading experts, a panel discussion, and a dedicated lightning-talk session for PhD students and postdoctoral researchers, the workshop aims to foster an interdisciplinary community developing the next generation of certifiable learning and control algorithms.

---

Invited Speakers

---

Organizers

• Peyman Mohajerin Esfahani — TU Delft and University of Toronto
• Lars Lindemann — ETH Zurich
• Dario Paccagnan — Imperial College London

---

Topics

The workshop focuses on learning and control under distribution shift, with topics including:

• Learning under distribution shift
• Distributionally robust control
• Safe learning-based control
• Certification and verification
• Statistical learning for control

---

Schedule