Hi, I'm Chenyang Lai
Interdisciplinary expertise in Robotics Control, Deep Learning, and Environmental AI, bridging intelligent systems and applied machine learning.
About Me
My introduction
I'm a AI and Robotics Developer with a background in Computer Science and Robotics, integrating ROS2-based automation with advanced AI frameworks such as PyTorch, Transformer architectures, and Diffusion Models to enhance intelligent perception and autonomous decision-making.
experience
projects
Skills
My technical levelProgramming
AI & Deep Learning
Reinforcement Learning
Robotics & Tools
Qualification
My personal journeyBEng in Computer Science And Technology
Guangdong University of Technology
2021 – 2025BSc in Computer Science
University of Exeter
2023 – 2025MSc in Robotic
University of Manchester
2025 – 2026 (expected)Portfolio
Most recent works
Oil Spill Detection via Change Detection
Designed an OSCD framework for bi-temporal SAR that generates realistic pre-spill counterparts with the TAHI pipeline—Dynamic Foreground Simulation, high-fidelity hybrid inpainting (PatchMatch + partial-conv U-Net), and radiometric/speckle alignment—so oil becomes the sole systematic difference. The resulting OSCD dataset (879 pairs) yields strong gains for CD backbones (e.g., FC-Conc/DDLNet) over differencing and single-frame segmentation, validating temporal supervision for robust maritime monitoring.
To develop a robust oil spill detection framework using bi-temporal SAR imagery and change detection, overcoming the limitations of single-frame segmentation for maritime environmental monitoring.
I independently designed and implemented the entire OSCD framework, including the novel TAHI pipeline (Dynamic Foreground Simulation, hybrid inpainting with PatchMatch and partial-conv U-Net, and radiometric/speckle alignment). I built the full OSCD dataset of 879 image pairs and conducted all experiments validating the approach against multiple CD backbones.
Media Gallery
Sustainable Development Platform
Led the development of Exeter Checkers, a gamified sustainable development web platform for the University of Exeter. Designed an interactive map-based system that promotes environmental awareness and encourages sustainable actions among students through real-world challenges, ranking systems, and achievement rewards.
To design and build a gamified sustainable development web platform for the University of Exeter that drives student engagement with environmental issues through interactive mapping and achievement systems.
I led the full-stack development of the platform, architecting the interactive map-based challenge system, designing the UI/UX, implementing the ranking and achievement reward mechanics, and coordinating the team to deliver a polished, deployable product.
Media Gallery
Credit Card Default Prediction
Designed and evaluated a deep neural network framework for credit default prediction, emphasizing the impact of activation functions and imbalance mitigation strategies. Leveraged Bayesian optimization and focal loss to improve model stability and class-wise performance, demonstrating strong generalization under k-fold cross-validation.
To investigate and improve deep learning approaches for credit card default prediction, with a focus on handling class imbalance and optimizing neural network architecture choices.
I designed the complete experimental framework, implemented the deep neural network with Bayesian hyperparameter optimization and focal loss, conducted extensive k-fold cross-validation experiments, and authored the final analysis comparing activation function effects and imbalance mitigation strategies.
Media Gallery