The Hong Kong University of Science and Technology (Guangzhou)
Linksome, Shenzhen
Wuhan University, Wuhan
Southern University of Science and Technology (SUSTech)
arXiv preprint
We propose MGDE, a novel method to detect error-inducing inputs based on mathematical guidance using the Newton-Raphson method with quadratic convergence properties. Our method identifies 89 bugs across 44 programs while being 6.4x faster than the baseline.
arXiv preprint
We propose PI-detector, a novel approach to computing floating-point errors effectively and efficiently by injecting small perturbations into the operands of individual atomic operations within the program and comparing the outcomes of the original program with the perturbed version.
Environmental Science & Technology
Employing an extensive flight tracking data set (Flightradar24), we assess the interflight variability in CO2 emissions from global aviation with unprecedented spatial and temporal granularity. We develop reduced-form models for over two hundred standard aircraft types that capture flight-to-flight variability.
arXiv preprint
We propose DELA, an approach based on the observation that most numerical errors stem from large condition numbers in atomic operations. DELA injects small perturbations into the results of individual atomic operations and compares the outcomes to detect errors.
Built a cross-platform Bilibili subscription companion that monitors creators, sends update notifications, and automatically caches videos to reduce link rot risk.
Built an AI-powered VS Code extension that converts natural-language find and replace requests into regex, then applies them directly to VS Code's built-in Find/Replace workflow.
Built a multimodal novel-to-video pipeline that uses LLM-based scene analysis, image generation, speech synthesis, and video composition to turn story text into animated short videos.
Calculated the canteen queue length from live monitoring data and visualized how the queue changed over time, attracting 30,000 visits in three months.
Used open-source LLMs such as Llama2-7B/13B to synthesize diverse JSON inputs, exercising 13 parsers with 100+ case types and surfacing 26 behavioral divergences.
Implemented a Minisys-compatible CPU that supports 32 core instructions, parses COE binaries, and interfaces with users through the Minisys development board.
Built a memetic and hybrid metaheuristic solver that reaches optimal CARP solutions on small and medium graphs within 180 seconds and delivers competitive performance on 255-vertex networks.
Developed a simple compiler (still has 15,000 lines) that translates C language files into Intermediate Representation (IR) and MIPS32 assembly. The compiler supports I/O operations, control flow, function calls, lexical analysis, syntax analysis, semantic analysis, and informative error messages.
The system lets players choose actions for their customizable characters, track HP, and engage in real-time multiplayer matches using miniSTM32 boards with accurate data transmission.
Implemented Monte Carlo search together with FAST alpha-beta pruning accelerated by bitwise operations to build a strong Reversed Reversi AI.
Processed 10 TB of Flightradar24 data covering 153 million flights to model aircraft-level CO2 emissions, explain 99.4% of emission variance, and highlight taxi optimization strategies that cut emissions by up to 20%.
Developed a visually engaging Othello experience with an interactive Swing interface and smooth in-game animations.