• Enable the synthesis of a circularly polarized 53-attosecond pulse in a single Helium atom response using Bayesian optimization. • GPU coding: Achieve a real-time computation time for a phase-sensitive spatial propagation PatchMatch with fractional tilt-filter for motion estimation in ultrasound imaging. It was 50 sec per frame on Matlab, and now 90 msec on GPU.

• Led the project in application of deep neural network to heart disease screening, including atrial fibrillation (AF) and myocardial infarction (MI) . • Collected over 500 long-term electrocardiograms from clinics, and improved the F1 score of identifying patients with AF and MI from 0.83 to 0.93. • Performed and supervised over data cleaning, signal segmentation, labelling, and unifying data format. Built 11-layers convolutional neural network from scratch to classify AF, MI, normal rhythms, and noise. • Initiated customer-oriented projects. Supervised the project of prescreening sleep apnea patients at Center of Sleep Disorder, Chang Gung Memorial Hospital Keelung Branch, Taiwan.

• Performed ab initio simulation by solving Schrodinger equation for a hydrogen atom rigorously using a pseudospectral method. • Applieid principal component analysis and diffusion maps to molecular dynamics to reveal the microscopic dynamic of water molecules. Presented in Physics informed Machine Learning workshop 2019. • Cooperation with experiment groups. Proposed an image processing method and apply maximum likelihood estimation to evaluate inhomogeneous gaps in large-Scale inhomogeneous fluorescence plasmonic silver chips. • Published 16 full journal articles (6 first authors) , including Nature Communication.

• Proposed a model for fast-varying instantaneous frequencies. • Proposed a time-varying window technique to cope with the artifacts aroused in linear-type time-frequency analysis. • Dimension reduction in molecular dynamics.