Zongyou Yang
GitHub: tongjiu123 | E-Mail: yzy0624@outlook.com | Phone: +44 07778418008
Education
University College London (UCL), London, UK
Master’s Degree in Computer Graphics, Vision and Imaging — 09/2024 – 09/2025
Thesis grade: A (Distinction) | Classification: Distinction
Core courses: Image Processing (72), Machine Vision (70.8), Computer Graphics (78), Machine Learning for Visual Computing (66), MSc Project (72)Beijing University of Post and Telecommunications (BUPT) & Queen Mary University of London (QMUL)
Bachelor’s Degree in Telecommunications Engineering with Management (Joint 4+0 Program) — 09/2020 – 07/2024
GPA: 3.71/4 (Top 6%, First Class)
Core courses: Digital Signal Processing (97), Principles of AI (92), Telecoms Systems (97), Image and Video Processing (94), Internet Protocols (95), Electromagnetic Fields and Waves (91), Computer Vision (83), Deep Learning Theory and Practice (98), Engineering Mathematics (93)
Publications
- Chaochen Hu, Ran Li, Chao Li, Hengshuo Miao, Zongyou Yang, Tengda Zhang. Big Data Analysis for Anti Money Laundering: a Case of Open Source Greenplum Application. WISA 2022.
- Haobo Yang, Shiyan Zhang, Zhuoyi Yang, Xinyu Zhang, Jilong Guo, Zongyou Yang, Jun Li. Eloss: An Interpretability Amplifier of 3D Object Detection Network for Intelligent Driving. arXiv:2409.00839.
- Zongyou Yang, Jonathan Loo. PyCAT4: A Hierarchical Vision Transformer-based Framework for 3D Human Pose Estimation. arXiv:2508.02806.
- Zongyou Yang, Yinghan Hou, Xiaokun Yang. Application of Hybrid Chain Storage Framework in Energy Trading and Carbon Asset Management. BIOTC 2026, Under Review.
- Yinghan Hou, Zongyou Yang. AI Agent for Reverse-Engineering Legacy Finite-Difference Code and Translating to Devito. arXiv:2601.18381.
- Patent: A Smoking Identification Technology Based on Deep Learning and Communication Big Data.
Research Experience
University College London — Master’s Thesis Project, supervised by Prof. Chris Xiaoxuan Lu (04/2025 – Present)
Hesitation and Detours: Spatiotemporal Digital Biomarkers for Alzheimer’s Disease- Developed a complete pipeline for AD behavioral recognition from GPS and video data, encompassing stop point detection, trajectory segmentation, optimal route comparison, and macro/micro-level behavioral metric extraction.
- Introduced Fréchet distance, detour/optimal/time ratios, direction consistency, and spatial entropy features via Google Maps optimal route alignment, establishing a two-level route comparison framework.
- Expanded modeling from logistic regression to RF/DT/SVM voting, XGBoost, MLP, and BiLSTM/Transformer combined with SMOTE, increasing F1 score from ≈0.60 to ≈0.83 under LOUO.
- Conducted SHAP and causal inference analyses (PSM/DML/T/S-Learner) to identify key actionable indicators.
Queen Mary University of London — Bachelor’s Thesis Project, supervised by Prof. Jonathan Loo (11/2023 – 05/2024)
PyCAT4: A Hierarchical Vision Transformer-based Framework for 3D Human Pose Estimation- Developed a novel human pose estimation model using coordinate attention and Swin-Transformer architecture.
- Designed a temporal feature fusion strategy and spatial pyramid module for multi-scale feature fusion.
- Achieved 100%–300% performance improvement on MPJPE over baseline CNN models on COCO and 3DPW datasets.
- Deployed real-time inference on Ubuntu 20.04 for live camera input applications.
Tsinghua University — Research Project, supervised by Prof. Xinyu Zhang (05/2022 – 06/2023)
Entropy Loss: An Interpretability Amplifier of 3D Object Detection Network for Intelligent Driving- Addressed the black-box challenge of 3D object detection in autonomous driving, focusing on improving model interpretability.
- Modeled intermediate network outputs as continuous random variables and applied KNN-based entropy estimation.
- Integrated Eloss into training as a complementary regularization term, enhancing both training efficiency and network interpretability.
BUPT — National “Challenge Cup” Project, supervised by Prof. Xiangdong You (07/2022 – 06/2023)
AI-Powered Smart Mini Photo Studio (“Snap Lover”)- Developed an AI-powered smart mirror using a Linux-based system with recording, voice recognition, and image processing.
- Implemented speech recognition using Baidu Speech Recognition API and red-eye removal algorithm.
- Extracted features from face/smiling-face datasets using Haar-like features and trained ML classifiers.
Tsinghua University — Research Project, supervised by Prof. Chao Li (11/2021 – 06/2022)
Big Data Analysis for Anti-Money Laundering: A Greenplum Application Case- Designed an efficient big data analysis solution leveraging Greenplum’s shared-nothing MPP architecture.
- Wrote SQL queries to process over 50 million records, imported data via gpfdist to evaluate query performance.
RLChina 2025 — Reinforcement Learning Workshop (09/2025 – 10/2025)
- Developed RL agents for autonomous decision-making in Tencent’s Honor of Kings 1v1 battle environment.
- Built an exploration-and-survival agent in the “Unlimited Valley” environment using reward-driven policy optimization.
- Applied advanced methods in DQN and actor-critic frameworks.
Work Experience
Peking University Wuhan AI Research Institute — LLM & Agent Researcher (10/2025 – Present)
- Led SFT of government domain LLMs based on DeepSeek-70B and Qwen3-32B. Achieved 26% improvement in structured output quality (content compliance scoring 9.90/10).
- Designed and deployed the JiangXia community intelligent agent system using a 1+3+5+N multi-agent architecture with RAG and local knowledge base.
- Built a complete validation pipeline for virtual digital humans covering asset creation, motion driving, and delivery.
- Developed a dual-agent system (learning companion + AI teaching assistant) using Baidu Qianfan AppBuilder.
TG0 Ltd. — AI Intern, SWEE Team, London, UK (05/2025 – 09/2025)
- Developed a 3D human pose estimation pipeline by fusing MediaPipe 2D keypoints with RealSense RGB-D and IMU data.
- Developed a deep learning motion capture system converting MediaPipe 3D keypoints to Rokoko joint angles, evolving from MLP to bidirectional LSTM with attention, reducing errors to within 5–20 degrees.
- Contributed to the Bentley PM Seats project and GP3 Smart Golf project.
China Mobile Communications Group Co., Ltd. — Intern, Big Data Department, Jinan, China (07/2024 – 08/2024)
- Researched privacy-preserving computing technologies and assessed feasibility in enterprise-level applications.
- Analyzed Migu Sports data using machine learning to identify high-conversion potential users.
- Built an anti-smuggling rule system and smoker database using privacy computing and telecom data standards.
- Co-authored two invention patent drafts on smoker identification technology.
4Paradigm — Intern, R&D Department, Beijing, China (07/2023 – 03/2024)
- Optimized workload scheduling in stream processing using time series prediction algorithms, enhancing OpenMLDB performance.
- Contributed to system-level optimization focused on reducing LLC cache misses through CPU and cache architecture analysis.
- Completed dynamic workload scheduling based on LSTF for ML databases.
- Integrated LSTF strategies from both statistical models and deep learning methods into production-grade solutions.
Extracurricular Activities
AI & ML Winter School, National University of Singapore (01/2022 – 02/2022)
Recognized as part of an “Outstanding Group” for teamwork and innovation.Song Qingling Foundation Tibet Support Program, Chaya County, Chamdo, Tibet — IT Volunteer (08/2020 – 09/2020)
Provided technical support for educational and community services in remote areas.
Selected Awards & Honors
- Outstanding Graduate, BUPT, 2024
- University Scholarships (Second/Third Prize), BUPT, 2020–2024
- Second Place, 19th CCF Innovation Design Competition & AgileSoft AML Rules Competition, 2022
- Third Prize, Blockchain Innovation Contest, 8th ICCIP, 2022
- National Second Prize, 33rd National Chemistry Olympiad, 2019
Skills & Languages
- Technical Skills: Python, C/C++, Java, SQL, Matlab
- Languages: Chinese (Native), English (IELTS 7.0)
- GRE: Verbal 156 | Quantitative 170 | Writing 3.5
- Hobbies: Fitness, Badminton, Traveling