Constraints · Verification · Guarantees

Learning With Constraints/Guarantees

This direction studies how constraints and guarantees can be incorporated into learning itself. Rather than checking desired properties only after training, the learning process can encode them through structured parameterizations, optimization layers, regularization, verification, or other principled mechanisms. The broader goal is to develop learning systems whose behavior is reliable by construction.

Learning With Constraints/Guarantees visual — Image: Miao et al., Learning Koopman Representations with Controllability Guarantees

Related publications

All publications

Designing neural controllers with optimality and stability guarantees by learning input-output dissipativity thumbnail

Designing neural controllers with optimality and stability guarantees by learning input-output dissipativity

H Wang*, K Miao*, DS Madeira, A Papachristodoulou

Automatica 191, 113120, 2026

arXiv Scholar Published

Learning Koopman Representations with Controllability Guarantees thumbnail

Learning Koopman Representations with Controllability Guarantees

K Miao, H Wang, X Ding, K Gatsis, A Krause, A Papachristodoulou

The Fourteenth International Conference on Learning Representations, 2026

Scholar Published

Opt-ODENet: Neural ODE controller design with differentiable optimization layers for safety and stability thumbnail

Opt-ODENet: Neural ODE controller design with differentiable optimization layers for safety and stability

K Miao, L Zhao, H Wang, K Gatsis, A Papachristodoulou

Proceedings of the 7th Annual Learning for Dynamics and Control Conference, PMLR 283:1217–1229, 2025

PDF PMLR BibTeX Scholar Published

How deep do we need: Accelerating training and inference of neural ODEs via control perspective thumbnail

How deep do we need: Accelerating training and inference of neural ODEs via control perspective

K Miao, K Gatsis

Proceedings of the 41st International Conference on Machine Learning 235 …, 2024

BibTeX Scholar Published