My name is Yuze (or Lucas), a third-year student in University of Toronto. I double
majored in mathematics and computer science, aiming for a PhD degree in the future. Currently, I am
interested in algebraic number theory and probabilistic graph theory. Also, I am keen on bridging
real-world problems with abstract theories by using machine learning.
"读万卷书,行万里路。 -- 司马迁 《史记》" This sentence is from a Chinese history book writen 2000 years ago,
meaning that True wisdom is born from the union of deep knowledge gained through study and profound
understanding earned from real-world experience. Travelling is a must of my every holiday.
Gaming also plays an important role in my life. I mainly play two types of game: sandbox and rpg. My
favourite sandbox game is Minecraft which accompanies me since 2012. Legend of Zelda: Breath of
the Wild is my best rpg game forever, still fascinating me time after time (I replay the whole
game at least 5 times).
Portfolio
Machine Learning-Driven NSI Forecasting for Toronto Neighborhoods
Developed predictive models for monthly neighborhood safety using 100K+ crime records, achieving 0.0527 RMSE
(best) through LSTM architecture that captures 12-month temporal patterns across 159 Toronto neighborhoods
Engineered Neighborhood Safety Index (NSI) using severity-weighted crime aggregation and temporal features
(lagged values, rolling averages) to transform raw incident data into actionable safety metrics
Compared 6 ML approaches (Ridge, KNN, FNN, LSTM, SARIMA) through systematic hyperparameter tuning; LSTM
achieved 82.95% R² while Ridge Regression provided interpretable coefficients for policy applications
Virtual Memory and File System Implementations for Operating System
May - July 2025
Tools Used: C, Pintos, Docker
Threads: Used Multilevel Queue Scheduling and Priority Donation to optimize Context Switching
User Programs: Implemented all basic system calls triggered by interrupts through Stack
Virtual Memory: Created Supplemental Page Table, Frame, Swap and Eviction Algorithm to support paging
File System: Chosed Multilevel Indexed File to support flexible file operation and implemented Directory
Sychronization: Tried both Lock and Semaphore to deal with different situations
Firefighters - Assembly Platform Game
April 2025
Tools Used: Assembly, Computer Organization, MARS
Move the firefighter to save injured people withoug touching the fire, and approach the exit as you can
Implemented Bitmap for display, MMIO Simulator for input
Actively used registers for efficient operation, and stack for function calling
Linux OS Monitoring Tool & FD Table Creator
April 2025
Tools Used: C, Linux, System Programming
Demonstrated real-time RAM usage, CPU usage by line charts and File Descriptor table
Captured information from Linux OS files and created diagrams, such as reading CPU usage time from "/proc/stat",
then calculating the real-time usage into 100%, and plotting onto the chart
Implemented pipes to enhance the performance by concurrency, and reinstalled signal handlers to fit the needs
Modular codes with the use of makefile, and detailed documentation including README and docstring
Fungi Image Classification & Denoising
March 2025
Tools Used: Python, PyTorch, CNN
Developed a classifier achieving 93.8% accuracy on a 5-class, 3,000+ image fungi dataset using GPU acceleration.
Implemented a denoising autoencoder with transposed convolutions to preprocess corrupted images.
Built a custom data pipeline with augmentation for memory-efficient loading and batch processing.
Performed noise analysis and statistical characterization to optimize preprocessing and enhance performance.
Engineered a smart sprinkler system using a KNN machine learning model (Python, OpenCV) to analyze lawn
images and enable targeted, water-efficient irrigation.
Achieved 80%+ accuracy in classifying lawn hydration levels by processing RGB image data and implementing
grid-based analysis.
Designed end-to-end system architecture from image capture to sprinkler control.