Controlling Dynamics of Complex Systems: Nonlinear Techniques vs Reinforcement Learning

复杂非线性系统吸引了当代科学许多不同领域的研究者的注意力。在理论和实验中广泛研究的一个重要现象是振荡网络中由于同步跃迁而出现的集体模式。这种集体模式可能是非常可取的，也可能是有害的。例如，假设神经元网络中的过度同步是癫痫和帕金森病中出现病理性节律的原因。该项目的一个主要目标是开发同步控制技术，将反馈控制与机器学习的现代子学科强化学习相结合。考虑到一种可能应用于医疗技术，称为脑深部电刺激，我们专注于基于脉冲动作应用的控制方案。非线性能够建设性和破坏性地影响问题的另一个领域是成像。控制噪声和信号之间的模式混合可以从根本上提高图像质量。在这里，我们计划使用同步控制技术来操纵成像系统中的空间动力学。

Complex nonlinear systems attract the attention of researchers from many different fields of contemporary science. One important phenomenon extensively studied theoretically and in experiments, is the emergence of a collective mode in oscillatory networks due to a synchronization transition. This collective mode can be either highly desirable or harmful. For example, it is hypothesized that excessive synchrony in neuronal networks is responsible for the emergence of pathological rhythms in epilepsies and in Parkinson’s disease. A major goal of the project is to develop techniques for synchronization control, combining feedback control with the modern sub-discipline of machine learning, the Reinforcement Learning. Having in mind a possible application to a medical technique, called Deep Brain Stimulation, we concentrate on control schemes based on the application of pulse action. Another area where nonlinearity is capable of affecting the problem both constructively and destructively is imaging. Controlling mode-mixing between noise and signal one can essentially enhance the image quality. Here, we plan to use the techniques for synchronization control to manipulate the spatial dynamics in an image-forming system.