Princeton University

• Predictive Model Leadership: Spearheaded the design and deployment of advanced predictive models tailored for fusion experiments, integrating state-of-the-art machine-learning techniques for real-time (10ms) control 

• Data Infrastructure Management: Orchestrated and finetuned a robust data infrastructure for time series, enabling effective Petabit-level data operations. Enhanced the data quality via techniques like up-sampling. 

• Innovative Data Synthesis: Devised attention-centric models that harmonize data from varied diagnostic tools, enhancing context and accuracy. Pioneered a cutting-edge attention convolution-based autoencoder, achieving a staggering 200x data compression while retaining vital details.

• Reservoir Computing (RCN) Implementation: Utilized ultra-fast RCN for precise channel reconstruction.

• Signal Processing Expertise: Delved deep into signal processing, emphasizing anomaly detection and classification within multichannel spectrograms, unveiling deeper data insights.

• Collaborative Project Approval: Championed and secured permissions to apply innovative methodologies in the nation's leading fusion research facilities, ensuring practical application and real-world testing of theoretical models. The work has been featured in 2 publications.

the University of Texas at Austin

 Large scale simulations, Data management, Data Visualization

• Simulation Excellence: Conducted extensive linear and nonlinear simulations with over 4M core hours

• Data Standardization: Addressed inconsistencies data by developing robust pre-processing tools. 

• Post-Simulation Data Extraction: Designed and refined an extensive suite of post-processing tools; handled vast data sets, with individual simulations generating data up to 30TB.

• Experimental Diagnostic Replication: Innovated a synthetic diagnostic tool mirroring experimental apparatus, strengthening the bridge between simulated outcomes and real-world results.

• Academic Contribution: The research were encapsulated in 4 published journal articles, 1 invited talk

A reduced model SLiM and Graphic User Interface (GUI) for such model

• Decade-long Mystery Solved: Developed the pioneering SLiM model using kinetic global theory for microtearing modes (MTM); the first-ever global MTM model that successfully demystified a decade-long unresolved phenomenon. This work is underscored by 4 journal publications and recognition in 2 invited talks.

• Optimization Mastery: Enhanced simulation speeds by a factor of 100 utilizing both GPU and CPU resources. Additional structural optimization strategies yielded even greater computational acceleration.

• Parallel Computing Achievement: Implemented parallel code suitable for the national computational center, a feat that would otherwise necessitate a staggering 15 years on a conventional PC.

Other

• A set of videos related to coding (python, C, FORTRAN), math (complex analysis, linear algebra, differential equations), physics (nonlinear dynamics, plasma physics), simulations are created to educate the new coming graduate students and interested undergraduate students.

• Mentored an undergraduate student in a semester-long individual study class on creating a parallel code for the reduced model SLiM and testing such code in Texas Advanced Computing Center (TACC). Taught the student on basic API and objective-oriented programming.

the University of Texas at Austin

• Over 200 students have been taught during the work as a teacher assistant and head of teacher assistants.

• As a complement to the course material, over 50 instruction videos are recorded for students’ consumption. It includes the experiment instruction, core concept review, mid-term/final review. Such video series soon became the official instruction videos series for this class 

University of Illinois at Urbana-Champaign 

Neutron electron dipole moment (nEDM) experiment

• Tested high power (3Watt) infrared laser and mounted for the polarized Helium 3.

• Designed and constructed a polarimeter using a surface mount high gain op-amp, a low noise environment is needed for a useable signal-noise ratio. The printed circuit board (PCB) was designed using KiCAD with reducing cross talk and impedance in mind.

• Designed and printed the mounts for the optics and circuits by the 3D printer using Sketchup. All the optics and the circuits are placed in the metal box to be shielded from the high-intensity radio-frequency wave which will be turned on during the experiment.

Monte Carlo study of Quark-Gluon Plasma using photon jet observables

• The lead (Pb) lead collision creates the heated quark-gluon plasma that provides insight into the early stage after the Big Bang.

• Performed Monte Carlo simulations using PYTHIA and JEWEL to compare with the Pb-Pb collision at 2.8TeV conducted in the European Organization for Nuclear Research (CERN) A prediction was made on 5TeV collision for the future experiments. the results have been presented in a poster at the American Physical Society (APS) Division of Nuclear Physics (DNP) annual meeting.

• Managed the dataset using ROOT to visualize the ratio of the photon and jet momentum to be compared to the experiment.