School of Science Department of Physics 74 Physics-Guided Data-Driven Modeling to Understand Complex Phenomena and to Solve Real-World Problems Supervisor: ZHANG Rui / PHYS Co-Supervisor: LI Sai Ping / PHYS Student: IP Yan Hei / COSC Course: UROP 1100, Fall UROP 2100, Summer In this project, we continue exploring the swarming behaviour of agents in a 2D space, specifically focusing on milling behaviour, which we can observe when agents move in a circular pattern, often seen in nature with flocks of birds or schools of fish. Understanding this behaviour can provide insights into how the dynamics of group movement are affected by self-decision-making. The main goal of this project is to train our agents to exhibit milling behaviour effectively, using the insights gained from real-life data. To investigate this, we use a combination of machine learning techniques. For reinforcement learning, we implement the Multi-Agent Proximal Policy Optimisation (MAPPO) algorithm, which helps agents learn optimal behaviours through trial and error. Additionally, we apply a Convolutional Neural Network (CNN) for supervised learning, which allows us to analyse and interpret real-life data related to agent movements. Physics-Guided Data-Driven Modeling to Understand Complex Phenomena and to Solve Real-World Problems Supervisor: ZHANG Rui / PHYS Co-Supervisor: LI Sai Ping / PHYS Student: KOO Ho Yin / PHYS Course: UROP 1100, Fall UROP 2100, Spring Swimming at low Reynolds number requires the breaking of time-reversal symmetry and friction anisotropy. The swimming pattern that can lead to the breaking of symmetry is interesting. To study the swimming mechanism, we may start from a simple physic models, using the Rotne-Prager-Yamakawa approximation or the resistive force theory, which involves only solving systems of linear equations. In this project, we investigate the physics behind microswimers, such as NG swimmers, Purcell rotator, Tsang model and Henky bar chain model. We also study the many reinforcement learning model such as Q-learning, A2C, PPO, DDPG and DQN. Physics-Guided Data-Driven Modeling to Understand Complex Phenomena and to Solve Real-World Problems Supervisor: ZHANG Rui / PHYS Co-Supervisor: LI Sai Ping / PHYS Student: MAO Yifan / DSCT Course: UROP 1000, Summer This report presents my learning and research experience on the project “Bacteria ratchet motors”. Bacteria, as one of the representative type of active particles, have their fascinating ability of providing a nonequilibrium system, exhibit remarkable potential for converting biological activity into useful mechanical work at microscopic scales. Building on experimental models by R. Di Leonardo, Andrey Sokolov and others who studied bacteria as propelling power for mechanical systems by using asymmetric ratchets to harness bacterial motion as propulsive force, after considering and incorporating basic bacteria collective motion features and statistics, I eventually developed a preliminary simulation code as a reproduction of their experiment.
RkJQdWJsaXNoZXIy NDk5Njg=