Neural Networks in Control and Dynamics" (abr. NNCD) develops a rigorous foundation for learning‑based modeling and control of dynamical systems. To this end, the course develops the mathematical foundations of neural ordinary differential equations (neural ODEs), partial differential equations (PDE-informed networks), and recurrent neural architectures including RNNs, LSTMs, and GRUs. Building on these fundamentals, the course culminates with attention-based architectures such as Transformers and their emerging role in modeling dynamics and decision-making.

Given below are the contents for the course:

1.  The course begins with neural ordinary differential equations (neural ODEs), linking architectures to numerical integrators and the adjoint method.
2. Neural PDE surrogates are treated via physics‑informed neural networks (PINNs) and neural operators, with attention to discretization and well‑posedness.
3. Recurrent models (RNN, LSTM, GRU) are presented as learned state‑space models for identification, forecasting, and feedback design.
4. Transformers are introduced for long‑horizon sequence modeling, decision‑making, and data‑efficient control.
5. Optimization for training covers constrained learning, regularization, adjoint gradients, and differentiable simulation.
6. Learning‑in‑the‑loop is addressed through model predictive control with learned dynamics and safety filters.

• Goodfellow, I., Bengio, Y., Courville, A.:  “Deep Learning”. Cambridge:  MIT Press, 2016.
• Chen, Ricky TQ, et al.: "Neural ordinary differential equations." Advances in neural information processing systems 31 (2018).

Prof. Dr.-Ing. Naim Bajcinca
naim.bajcinca(at)mv.uni-kl.de
+49 631/205-3230
Gebäude 42, Raum 262
Sprechstunde: nach Vereinbarung

Übungsleiter

Dr. Mohammad Al Khatib

KIS entry

Oral Exam: 15-30 Min.
Date: 27.02.2026
Time: 14:00 – 16:30 Uhr
Location: Building: 42-260
Credit Points: 3ECTS
KIS entry

System and control theory, 

linear algebra,

python programming

OLAT