Symbolic Kolmogorov-Arnold Networks as a Constant-Time Alternative to Linear Time-Varying Model Predictive Control for Embedded Control of Non-Minimum Phase Systems
Adilkhan Salkimbayev
Overview
The project addresses a gap present in existing control architectures: while Model Predictive Control represents a rigorous paradigm capable of robust and optimal control, it is also computationally demanding. The paper proposes a symbolic distillation of the MPC control law into a lightweight polynomial via Kolmogorov-Arnold Networks (KANs), yielding increased performance by up to 5 orders of magnitude.
My role
The project was designed and architected by me from start to finish.
Approach
Existing literature was used to design the MPC control law. The distillation was performed by me, providing a basis for future research.
Results
Links
Remark
Even though this project is minted with a later date, it had been finished earlier and had been originally sent to IEEE ICCA 2026. Upon feedback, it was instead re-routed to Engineering Applications of Artificial Intelligence and is currently under review there.