Berlin 2024 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

DY: Fachverband Dynamik und Statistische Physik

DY 34: Poster: Machine Learning, Data Science, and Reservoir Computing

DY 34.13: Poster

Wednesday, March 20, 2024, 15:00–18:00, Poster C

Feedback Controlled Microscopy Using Machine Learning — •M Asif Hasan and Frank Cichos — Faculty of Physics and Earth System Sciences, Leipzig University, Linnéstraße 5, Leipzig, Germany

Feedback control is crucial in stabilizing unstable states, evident in living organisms to regulate system functions and in technological applications like quantum state control. The combination with machine learning offers novel approaches to probe and manipulate complex physical or chemical processes, where machine learning algorithms determine the control strategy for inducing specific physical or chemical perturbations in a microscopic system. Our project investigates a more cohesive approach to feedback-controlled microscopy, particularly in steering active microparticles amidst complex, noisy environments. To this end, we integrate a microscope and laser steering system with real-time particle detection and machine learning enabled feedback algorithm, specifically, the Actor-Critic Reinforcement Learning (ACRL) approach. We show that the AI agent can navigate the active particles and complex mixtures of passive particles to a target state with high precision, amidst environmental uncertainties such as Brownian motion and flow fields. With a multi-agent real-time learning design, we focus on navigating, adapting, and optimizing behaviors under fluctuating conditions, enabling the agents to proficiently interpret sensory data and learn optimal response policies in continuous action spaces. This study therefore paves the way to develop a universal Actor-Critic Reinforcement Learning multi-agent system enabling high-precision control in noisy settings of various fluidic scenarios.

Keywords: Feedback controlled microscopy; Actor-Critic Reinforcement Learning; Machine Learning