报告题目：Implementing Machine Learning to Physical Experiments
报告题目：吴亚东 博士 清华大学高等研究院
Recently, machine learning has become a very popular method for solving physical problem. Due to its powerful ability, it’s wildly used for regression, classification, controlling et. al.. In this talk, I will talk two projects about implementing machine learning methods to the physical experiments. The first one is using independent component analysis(ICA) method to analyze experimental data. For the BEC trapped in a harmonic trap, there are many kinds of excitations which are mixed. Thus, it is a tough task to know many modes and their frequencies. We find that ICA can recognize different collective modes and analyze the exciting frequencies. The second one is using machine learning method to find a best tune parameters in experiment. For unitary gases in a time dependent harmonic trap with frequency omega(t)=1/lambda t, the scale invariant symmetry is preserved and with time evolution, cloud has logarithmic plateaus. It’s called Efimovian expansion. We find the active learning can find a procedure omega(t) to release BEC cloud in a harmonic trap from absolute initial frequency omega_i to final frequency omega_f as fast as possible, which is much faster than the adiabatic evolution. As a result, the excitations, like breathing mode, can be suppressed. The procedure optimized by active learning can be saved for the Efimovian expansion experiment.
Yadong Wu now is a PhD student in Institute for Advanced Study, Tsinghua University. She graduated from Department of Physics, Renmin University of China. Her research field includes machine learning and cold atom physics.