Jui-Yang Hsu
Software Engineer
Research Scientist
Biography
I am a software engineer at Google Taiwan with 3+ years of experience in software development and 8+ years of experience of hands-on applied machine learning in speech/language processing, mobile testing, and edgeAI.
As a software engineer, I specialize in translating ambiguous stakeholder requirements into clearly-defined and actionable tasks. I bring a strong analytical mindset and a focus on system optimization, enabling the team to troubleshoot and resolve issues across systems efficiently. I am committed to upholding best practices in software quality, scalability, performance, testability, and adherence to style guidelines.
As a researcher, I am driven by curiosity and the desire to tackle real-world challenges through innovative algorithms/models with a tack record of success including modern AI model performance optimization on SoC, keyboard path recognition on iOS and more.
Beyond work, I’m a passionate sports enthusiast and amateur athlete, with interests ranging from table tennis and hiking to marathon running and strength training.
- Software Architecture
- ML System Design
- EdgeAI Model Optimization
-
MSE in Computer Science, 2018 - 2021
National Taiwan University
-
Exchange Student in CSC, 2017 - 2018
Kungliga Tekniska högskolan
-
BSE in Electrical Engineering, 2013 - 2018
National Taiwan University
Experience
- Building infrastructure to revolutionize software development life cycle.
- Architect and contributor of the in-house compiler auto-optimization toolkit
- Proposed, implemented, and maintained Prefect-based distributed computing platform for heterogeneous devices (host, various generations of smartphone platforms)
- Proposed algorithms for compiler auto-optimization,
boosted the inference speed in 20 ETHZ AI-benchmark models (out of 38) - Inventor of 4 patents in compiler auto-optimization, efficient and scenario-aware network architecture search (NAS)
- Improved customers AI-model efficiency on time and power via team-developed toolkit
Compiler Auto-Optimization System Design Clean Code Architecture Search Quantization-Aware
- Proposed and refactored model training to Pytorch Lightning for faster development and easier maintenance
- Migrated model training to the official AzureML training pipeline
- Implemented unified multi-vertical document understanding model
for variaous kinds of documents’ named entity recognition (NER)
As the graduate student
- Graduate Researcher in Speech Processing & Machine Learning Lab,
supervised by Prof. Hung-Yi Lee - Researched on low-resource speech recognition and improved the system with gradient-based meta-learning and transfer learning algorithms
- Lead TA of Deep Learning for Human Language Processing (Spring 2020) [CommE5054]
As the network adminstrator
- Managed a 10+ nodes, 20+ GPU Slurm-based computing cluster
- Incorporated netdata to the workstation for real-time monitoring
- Developed automatic health check to improve user/administrator experience
- Researched on generative modeling to develop algorithms enhancing user experience during keyboard usage
- The research results have been incorporated in iOS 13 and published as US patent
- Proposed and built the interface between Kaldi & Tensorflow,
migrating acoustic modeling part to Tensorflow for faster development - Researched on end-to-end speech recognition based on alignment-free algorithm (CTC)
- Reduced 3% CER on the corpus held by DRC, comparable to the original system
- Undergraduate Researcher in Speech Processing & Machine Learning Lab
- Proposed hierarchical attention-based model for TOEFL Listening Comprehension Test by machine
- Researched on self-supervised audio embeddings
- TA of Machine Learning → EE5184 designed assignment
Publications
Contact
- sunprince12014@gmail.com
- R531, EE Building Ⅱ, Dept. of Electrical Engineering, NTU <br/> No. 1, Sec. 4, Roosevelt Rd., Da’an Dist., Taipei City, 106319