Biography
My strongest research area is in machine learning, with a specific focus on classification and detection problems. This includes working on attack detection and the development of advanced techniques like generative adversarial networks (GANs) to efficiently generate adversarial examples for the adversarial training. My work primarily uses feature engineering and data quality improvement techniques.
Contact me at: rliu2{at}umbc{dot}com.
Download my resumé .
- Machine Learning
- Cybersecurity
- Time Series Model
- Back-End Development
PhD in Computer Science, 2024
University of Maryland, Baltimore County
MS in Computer Science
University of California, Santa Barbara
MS in Information Security
Johns Hopkins University
Experience
Responsibilities include:
- Designed machine learning predictive model using PyTorch for competitors identification through Hidden Markov Model, Convolutional Neural Network and Dynamic Time Warping, improving detection accuracy from 46% to 67%.
- Developed RESTful APIs for the production system using Java Spring Framework, and implemented a lightweight version algorithms using MySQL, benefiting over 30,000 clients.
- Designed statistical model to predict errors in forecasting system through Random Forest and ARIMA, decreasing the mean absolute percentage error (MAPE) from 0.17 to 0.04, contributing to over $12 million revenue.
- Designed clustering algorithms utilizing GNN and GAN with TensorFlow. Refactored the base code using Java functional programming, reducing running time from approximately 8 days to around 3 hours.
Responsibilities include:
- Designed and implemented new I/O APIs optimizations using C++ that were used by all teams in the AWS Aurora division, reducing the IOPS (Input/Output Operations Per Second) of the AWS Aurora over 80%.
- Designed a new batch data structure for I/O submission, reducing system calls by a factor of 4,096.
- Refactored code to eliminate lock dependencies using multithreading techniques, reducing I/O latency by 77%.
Responsibilities include:
- Led 12 penetration tests on web and mobile applications using Burp Suite, NMAP, Windows shell, and Python. Delivered comprehensive risk assessments and recommended solutions to managers and stakeholders.
- Led code review processes before application releases, offering in-depth risk assessments and proposing technical solutions to enhance security and programming standards.
- Developed statistical models using Hidden Markov Model and Exponential Smoothing to detect transaction fraud.
Projects
- Designed machine learning predictive model using PyTorch for competitors identification through Hidden Markov Model, Convolutional Neural Network and Dynamic Time Warping, improving detection accuracy from 42% to 67%.
- Developed RESTful APIs for the production system using Java Spring Framework, and implemented a lightweight version algorithms using MySQL, benefiting over 30,000 clients.
- Designed statistical model to predict errors in forecasting system through Random Forest and ARIMA, decreasing the mean absolute percentage error (MAPE) from 0.17 to 0.04, contributing to over $12 million revenue.
- Developed C++ I/O APIs for AWS Aurora, achieving an 80% reduction in IOPS across all teams.
- Created a batch data structure for I/O, cutting system calls by 4,096 times.
- Enhanced code using multithreading to remove lock dependencies, lowering I/O latency by 77%.
[Read More]
A fault-tolerant distributed file system that supports concurrent read/write operations using AWS, Java RMI, and the Primary protocol. The system has good scalability which can handle 10,000 requests with a maximum of 10,000 requests running concurrently.
[Read More]
