I am a PhD Fellow in the Machine Learning Group at the SFI Visual Intelligence centre, UiT – The Arctic University of Norway (March 2026–present). My doctoral research focuses on weather forecasting models, neural operator learning, and continual learning — aiming to develop robust, data-efficient deep learning methods for complex physical and spatio-temporal systems.
I hold an MSc in Computer Science and Engineering from the Technical University of Denmark (2023–2025), where I graduated with the highest grade (12/12) on my thesis on IoT honeypot security. I also hold a BSc in Electronic Information Science & Technology from Beijing Information Science & Technology University (Top 5% of cohort, GPA 3.70/5.0).
Before academia, I spent three years as an Embedded Software Engineer at Beijing Highlander Digital Technology, where I developed maritime navigation and voyage data recording systems, and held a CN patent on ship model parameter estimation. My background bridges low-level embedded systems with modern machine learning — giving me a distinctive perspective on applying AI to real-world physical environments.
Developing and adapting deep learning models for numerical weather prediction, exploring hybrid physics-informed and data-driven approaches to improve forecast accuracy and generalisation.
Investigating operator learning frameworks (e.g., FNO, DeepONet) for learning solution operators of PDEs, with applications in climate and fluid dynamics simulation.
Designing machine learning systems that can incrementally acquire knowledge from non-stationary data streams without catastrophic forgetting — critical for adaptive forecasting systems.
Making Vision Transformer-based foundation models self-explainable via keypoint counting classifiers — turning opaque ViTs into interpretable, human-communicable decision processes.
Honeypot architectures and psychological deception strategies for hardening IoT infrastructure against cyber threats, with data-driven dashboards for attacker behavioural analysis.
Real-time embedded software for maritime navigation, heading control, and voyage data recording — bridging sensor-level signal processing with system-level reliability requirements.
LinkedIn