Gradient-Based Adaptive Prediction and Control for Nonlinear Dynamical Systems

Researchers have developed a novel gradient-based adaptive prediction and control framework for nonlinear stochastic dynamical systems. The method works under a weak convexity condition, accommodating common nonlinearities like ReLU and sigmoid functions. Crucially, it achieves global convergence without requiring persistent excitation of the data and provides explicit asymptotic performance rates. The approach was validated on a real-world judicial sentencing dataset, demonstrating practical effectiveness for complex, real-time control applications where data is irregular.