Xinyi (Judy) Liang

858-***-**** ad2zcz@r.postjobfree.com 9500 Gilman Drive, Mail Code 0018, La Jolla, CA 92093

Professional Experience

Data Scientist, KPMG

Shanghai, CN

full-time intern, Risk Consulting Department

Jan 2022 - Mar 2022

Developed ETL pipelines by MySQL based on the annual reports of potential investment companies of the project from the past four years; Built interactive Power BI dashboards to visualize these companies’ distribution of expenditures and revenues.

Collected negative information of each target company from the Internet through Python crawler as risk reference indicators.

Analyzed the accounting audit processes and extracted key indicator data of the relevant companies by R functions (ANOVA correlation coefficient, Lasso Regression), assisted the team in detecting high-risk processes, and conducted horizontal and vertical comparisons to form a risk measurement report.

Designed and implemented A/B testing to determine the risk level of each company and gave the final decision of investment.

Research Assistant, Shenzhen Institute of Big Data

Shenzhen, CN

part-time intern, Application of RL to Radar Emission - Supervisor: Dr. Wenqiang Pu

Jul 2021 - Dec 2022

Used pytorch and Tensorflow to construct a neural network outside model-free Reinforcement Learning algorithms and trained the network by minimizing kl-divergence, called mb-mf algorithm. Mb-mf algorithm increased radar’s learning rate by 67.7% compared to existing radar training algorithms and improved the win rate of radar against jammer from 70% to 75%.

Formed a network ensemble using pre-trained networks on classical interference strategies and random networks with different initial weights, called me-mf algorithm, to further accelerated radar’s learning rate and increased the stability of radar performance tested in the real adversarial environment by 80%.

Research Assistant, Shenzhen Institute of Artificial Intelligence and Robotics for Society

Shenzhen, CN

part-time intern, Application of Computer Vision to Science and Arts - Supervisor: Dr. Ning Ding

Mar 2022 - Mar 2023

Worked for the interactive media artwork Special You in the Crowd. Used yolov7 to train a pedestrian detection model and integrated ByteTrack and SORT for tracking. Applied SparseInst to conduct instance segmentation on pedestrians.

Worked for the creation of Electric Power Inspection Robot. Trained a meter detection model using darknet deep learning network. Constructed a tracking system using CSRT on OpenCV to locate the meter accurately and adjust its camera precisely to read electric power. The robots have been in use to take over human’s mission in high voltage region.

Research Experience

The ScholarNet and AI Supervisor in Materials Science Research

Shenzhen, CN

Robotics and Artificial Intelligence Laboratory, Focus: NLP

Jul 2023 - Aug 2023

Fine-tuned MatSciBERT for corpus feature extraction. Extracted key concepts using pre-trained materials science word embeddings, and applied PCA for dimensionality reduction. Measured keyword importance using cosine similarity.

Executed node embedding on concatenated vector. Applied graph convolutional operations to generate tensor network states.

Constructed tensor network using Matrix Product State (MPS) to measure the similarity between provided article and ScholarNet reference articles for novelty and originality’s evaluation.

Time Series Prediction in Horse Racing Gambling

May 2023

Applied STL to decompose the time series, revealing insights into trend and seasonality patterns.

Identified optimal parameters for the ARIMA model based on ACF and PACF plots. Fitted the ARIMA model to the transformed regular-spaced data. Achieved a final normalized MSE of 0.3358 on the test set.

Constructed input feature sequences using past bets, daily race counts, true winning probabilities and fed them into LSTM model. The model significantly reduced MSE by 15%.

Education

University of California San Diego 4.0/4.0

San Diego, CA, USA

MSc. in Electrical and Computer Engineering (Stream: Machine Learning & Data Science)

Sep 2023 – Jun 2025 (expected)

The Chinese University of Hong Kong, Shenzhen

Shenzhen, CN

BSc. in Statistics (Stream: Data Science)

Sep 2019 - Jun 2023

Honors & Awards: Academic Scholarship: Class B 2021-22 (5/278), Dean’s List 2021-22, Outstanding Student Assistant 2019-20

Relevant Coursework: Data Structures, Parallel Programming, Techniques for Data Mining, Machine Learning, Data and Knowledge iManagement, Deep Learning and Applications, Numerical Methods, Time Series, Stochastic Processes

Skills & Interests

Technical Skills: Python (NumPy, pandas, pytorch, tensorflow, Matplotlib, SciKit-Learn, scrapy); R (ggplot2, dplyr, caret); SQL (MySQL), C++ (STL); Java; Hadoop; Spark; MATLAB; Advanced Excel; LaTex; Markdown; Visualization (Tableau, BowerBI)

Languages: Chinese Mandarin (Native), Chinese Shanghainese (Native), English (Professional)

Interests: Photography, Basketball, HipHop, Orienteering

Contact this candidate