Inhalt des Dokuments
Control in machine learning
Held by Dr. Markus Abel (Universität Potsdam)
29.04.2014, 16:00 Uhr
EW731
Abstract:
We present a new approach for the control of systems with complex dynamics: In order to determine the optimal control function we apply a machine learning algorithm, genetic programming. With this method and a clever-chosen cost function we are able to determine fully nonlinear control laws in a systematic way. We apply our algorithm to a turbulent mixing layer, and the Lorenz system in a periodic state with the goal to achieve maximum mixing, characterized by the Lyapunov exponents of the system. Further, we control coupled nonlinear oscillators to hold them in a stationary state.