Shiyuan Feng

- Developed a composed image retrieval system based on Vision-Language Models for facial data, using both image and text as references, significantly improving the accuracy of retrieval and addressing the challenge of selecting images with specific attributes. - Created an innovative dataset tailored for retrieving complex facial attributes and designed new evaluation metrics to measure retrieval performance at the attribute level.

- Developed an object detection model for components detection on printed circuit board based on YOLOv7 from scratch, handling data processing, model training, model quantization, and testing. - Introduced oriented bounding boxes to predict the slant angles of components with poor printing quality, experimenting with MMRotate, Rotated RetinaNet, and YOLO with Oriented Bounding Box frameworks.

Project of Commodity Futures Time Series Forecasting - Built a feasible and effective NI futures time series forecasting system based on LightGBM and performed data cleaning and feature engineering to select effective features using data analysis tools like Scikit-learn and Pandas. - Built an online version for real-time trading and demonstrated stable results.

Master of Science in Applied Computing, AI concentration. Ongoing courses: Software Engineering for Machine Learning, Computational Imaging, Neural Networks and Deep Learning.

-Graduated with Honours -Courses cover extensive knowledge in programming, signal processing, information theory, artificial intelligence and data science -Worked as a researcher in several machine learning projects