2023 - Present Ph.D. Student, Computer Science, Arizona State University Tempe, AZ
2021 M.S., Electrical & Computer Engineering, University of Washington Seattle, WA
2019 B.S. with honors, Physics, China’s Top-notch Undergraduate Training Program, Lanzhou University Lanzhou, China
2017 Exchange Student, University of California, Berkeley Berkeley, CA
ML Security and Fairness / Perception in Robotics / Probability and Random Processes / Convex Optimization / Computer Vision / Statistical Learning / Digital Image Processing / Artificial Intelligence / Natural Language Processing
Programming Languages: Python, C/C++, MATLAB, Verilog/System Verilog, FORTRAN
Platforms/Frameworks: PyTroch, Tensorflow, OpenCV, FPGA(Quartus), Virtuoso Cadence, Git, LATEX, LabView
Sept. 2021 – Nov. 2022
Machine Learning Engineer, Specialized in Computational Photography Ernst Leitz Labs | Leica Camera AG Bellevue, WA
• Investigate and develop deep learning based method to replace and optimize mobile camera Image Signal Processor (ISP) pipeline
• Participate in Leitz Looks Engine project: develop image rendering engine and depth estimation algorithms
• Integrate various depth estimation algorithms such as RAFT-Stereo and MiDaS onto next generation Qualcomm Snapdragon chipset, and perform speed/power optimizing and performance profiling
Arizona State University
Advisor: Xusheng Xiao, Associate Professor
Large Language Model (LLM) Powered Description Synthesis for Sensitive Behaviors in Mobile Apps∗ Design a unified framework using multi-granularity program data, program analysis tools, and multiple LLMtechniques to auto-generate descriptions for Android app privacy calls∗ Employ static program analysis using Soot to extract multi-dimension information from a large number of Androidapps for LLM processing∗ Establish the industry’s first dataset merging program info, privacy calls, activity descriptions, and keywords.Dataset under preparation∗ Work still in progressArizona State University
Advisor: Chaowei Xiao, Assistant Professor
Adversarial Demonstration Attacks on Large Language Models
∗ Pioneered the analysis of a novel threat model for LLM in-context learning, allowing attacks via demonstration
modifications
∗ Achieved high attack success on leading LLMs, including GPT2-XL and Llama, across multiple tasks
∗ Designed experimental methods to evaluate the quality of attacked samples through linguistic/semantic analysis,
model perplexity, and user analysis. Highlighting the stealthiness of the attack
∗ Paper under submission to top NLP conference. ArxivInformation Processing Lab | University of Washington, Seattle
Advisor: Jenq-Neng Hwang, Professor
Multi-camera Multiple People Tracking, ICCV2021 Workshop
∗ Used FairMOT-based single-camera multi-object tracking algorithm for single camera tracking
∗ Introduced a combined foot point estimation method, which integrated Keypoint-RCNN and statistical method
∗ Applied an association method to generate multi-camera tracking result, which includes three steps, frame-by-frame object association, overall track association and associated tracks merge
∗ The final result ranked third place in track2 and fifth place in track1 among more than seventy participants
Pose-Flow Multi-Object Tracking and Human Pose Estimation
∗ Developed Joint-Flow: an offline human pose estimation and tracking method combined temporal information
∗ Proposed a tri-branch architecture, contains two single frame human pose estimation branches, top-down and bottom-up, and a joint-flow branch
∗ Achieved great performance on PoseTrack dataset as 78.70 mAP and 63.57 MOTA
Earlier Research Experiences in Physics Domain
Dec. 2018 – May 2019Undergraduate Thesis | Electronic Materials Lab | Lanzhou University
Advisor: Deyan He, Professor
Application of MOF based metal oxide & sulfide in energy storage
∗ Explored a three-in-one method using Zeolitic Imidazolate Frameworks-67 as the precursor to synthesize oxide, sulfide and carbide as electrode materials, and reported excellent electrochemical performance
Functional Materials Lab | Lanzhou University
Advisor: Jiachi Zhang, Professor
Persistence luminescence of bi-phase & tri-phase solid solution
∗ Synthesized high-quality bi-phase and tri-phase solid solutions, and revealed deep-trap level and weakened afterglow Optical information storage materials in optical logic devices
∗ Revealed a good read-in ability, high read-out stability of LuAG under different excitation conditions, and designed several optical logic gates based on that material
Plasma & Metal Laboratory | Chinese Academy of Sciences
Advisor: Guangan Zhang, Professor
Controllable dispersion of Cu and Ni nanoparticles in lubricating oil
∗ Investigated the influence of surface modification on the dispersion of nanoparticles in lubricating oil
3rd place in Multi-camera Multiple People Tracking Challenge Camera View Track, ICCV 2021 Workshop 2021
Excellent students graduated with highest honors(top 5% in Lanzhou University) 2019
Excellent undergraduate theses of Lanzhou University (top 5% in Lanzhou University) 2019
Outstanding student leaders (top 2% in Cuiying Honors College with excellent leadership) 2017
Academic scholarship for outstanding students (top 10% with outstanding academic performance) 2017
Selected to China’s Top-notch Undergraduate Training Program 2016-2019