Kaohsiung City, Taiwan
He is a data science engineer with a mathematical and statistical background. He is experienced at understanding business logic, realizing them with programing language, and helping clients achieve their goals with a data-driven approach. Gary is not a typical data scientist: he is open-minded to be a data engineer or backend engineer, or even project manager. Any position creates an impact upon people, the market, and the world is interesting to him. And he will seize any opportunity to create value.
Jun 2021 - Present
MIT visiting Engineer program
- Research in close collaboration with Prof. John Guttag and his lab.
- Focus on test-time augmentation in segmentation problem.
- Introduce MIT's state-of-the-art techniques to Wistron.
- Build a bridge that links others in Wistron and the MIT CSAIL.
Sep 2018 - May 2021
Manufacture scheduling optimization
- client: a well-known plastic corporation
- Join this one-year project 3 months before it closes. Yet, fully understand the complicated business knowledge in two weeks.
- Rewrite code which handles the unsolved problem: (1) equalize machines' loading (2) algorithm's efficiency (a couple of hours -> 30 sec at most)
- Effectively assist production schedulers: it originally took two employees 8 hours a day to do this job.
- Introduce design pattern to development team, enhance codes' flexibility.
EIC syndrome (breast cancer) detection
- Build an AI model training platform (using Django), allowing domain users to train models without coding.
- Reach 0.83 AUC, which a physician might not achieve.
- Cooperate with domain experts, acquire medical knowledge from physicians and offer them mathematical insights.
Laptop keyboard material recommendation system
- Apply association rule to avoid some material combination with a high defect rate.
- Offer backend engineers a containerized API, fulfilling a comprehensive digital transformation service.
- Have frequent conversations with clients, understand business logic and needs, convince people based on data and statistical intuition.
- Reduce manufacturing costs (10 million NTD/month) by proper disposal of material waste.
Data augmentation with deformation
- Apply MIT's SOTA method: Voxelmorph to augment manufacturing data.
- Patent under review (Taiwan, China, USA and India)
Sep 2017 - Dec 2017
Integrated a series of numerical methods and machine learning methods and delivered a solution that detects defective products based on signal data.
Jul 2017 - Aug 2017
Focused on discovering dog tools (problematic machine) by building statistical models and defining a series of criteria