Shu-Yu Lin
"LIFE IS LIKE CODING, TRIAL AND ERROR."
Experience
Machine Learning Engineer Intern
VIA Technologies Inc.
Taipei, Taiwan
- Developed a key-feature extracting system with a unified vision-language model, capturing 10+ coarse-grained features in driving images, replacing 5+ specialized models, greatly reducing memory occupation on limited devices.
- Achieved real-time performance of 21 frames per second in M810, an edge device powered by a Qualcomm chip, utilizing ONNX and Qualcomm SNPE tools to transition, shrink and quantize model for faster inference speed.
- Built automated data annotation pipeline to curate 100k+ labeled images, boosting labeling efficiency by 200%+.
Undergraduate Student Researcher
Web Intelligence & Data Mining Lab, NCU
Taoyuan, Taiwan
- Designed a transformer-based model for forecasting of global total electron contents (GTEC), a critical indicator of the space environment, achieving state-of-the-art global predictions, advancing the understanding of neural-network applications on spatiotemporal data and resolving 4 hours latency issue in real-time GTEC data caused by transmission and computing with an error rate less than 5%.
- Devised transformer-based hierarchical model for rhetorical role labeling from legal judgments, a sentence-level classification task, introducing a novel BERT fine-tuning method for hierarchical semantic structure.
- Secured scholarship from National Science and Technology Council, the highest research and technology authority in Taiwan, initializing research project overviewing special tempo prediction methodologies.
Projects
KazumaSpeak - Japanese Learning App
Developed Android application implementing communication-first learning methodology with pronunciation practice, grammar discovery, and reverse translation features. Integrated speech recognition APIs, audio processing, Room database, and location services using MVVM architecture.
Autonomous Self-Driving Wheelchair
Led hardware integration of LiDAR and camera systems, developing ROS-based sensor fusion architecture and URDF models for autonomous wheelchair platforms. Conducted comparative analysis of LiDAR odometry algorithms, successfully implementing optimal indoor navigation.
Autonomous Robot Car
Built an autonomous robot car capable of navigating without collision, integrating localization, control, and planning modules, implementing particle filter for localization, PID and MPC for control, and Lazy A* for path planning.
GTEC Forecasting
Predict Global Total Electron Content. Paper Accepted as publication at Space Weather [Paper #2023SW003579RR]
2048 Auto Player
Develop self-play machine for unlimited 2048. Use Gym framework and PyGame visualization
Soccer Tracking
Track soccer game. Dataset from SoccerNet. YOLOv8 for object detection. StrongSORT for tracking.
Education
Master of Science, Electrical and Computer Engineering
University of Washington
Seattle, WA, USA
- Focus on Machine Learning and Computer Vision
- Coursework in Advanced Algorithms, Deep Learning, and Signal Processing
- Graduate Research Assistant
Bachelor of Science, Computer Science and Information Engineering
National Central University
Taoyuan, Taiwan
- Graduated with Honors
- Relevant Coursework: Data Structures, Algorithms, Machine Learning, Computer Networks
- Undergraduate Research in Space Weather Prediction
Awards
- Undergraduate Research Fellowship, National Science & Technology Council, TW
- 2nd Place, EECS College Special Topic Competition
- 3rd Place, CSIE Department Special Topic Competition