About Me

Yong-Cheng Liaw is currently pursuing a master's degree at the Institute of Artificial Intelligence Innovation, National Yang Ming Chiao Tung University, starting from September 2023. His research focuses on the system optimization of deep neural networks, as well as research in the context of non-volatile memory attributes such as NAND Flash Memory and Skyrmion Racetrack Memory (SK-RM).
Publications
- , Shuo-Han Chen and Yu-Pei Liang, "Reinforcement Learning-Based Read Performance Throttling to Enhance Lifetime of 3D NAND SSD," 13th IEEE Non-Volatile Memory Systems and Applications Symposium (NVMSA), Sokcho, Korea, August 21-23, 2024.
- , Shuo-Han Chen, Hsin-Yun Su, "Lowering the Number of Live-Page Copies on Solid State Drives through Trim-Assisted Space Allocation," IEEE International SoC Conference (ISOCC), Jeju Island, Korea, Oct. 25-28, 2023.
- Yu-Shiang Tsai, Shuo-Han Chen, , Cheng-Yueh Wu, "Exploring Hot/Cold Data Separation for Garbage Collection Efficiency Enhancement on OCSSDs," IEEE International SoC Conference (ISOCC), Jeju Island, Korea, Oct. 25-28, 2023.
- , Shuo-Han Chen, Yuan-Hao Chang, Yu-Pei Liang, "Sky-NN: Enabling Efficient Neural Network Data Processing with Skyrmion Racetrack Memory," 2023 ACM/IEEE International Symposium on Low Power Electronics and Design (ISLPED 2023), Aug. 7-8, 2023, Vienna, Austria. (Top Conference)
Academic Activities
- Oral Presentation at the IEEE Non-Volatile Memory Systems and Applications Symposium (NVMSA), Sokcho, Korea, August 21-23, 2024.
- Oral Presentation at the IEEE International System-on-Chip Conference (ISOCC), Jeju Island, South Korea, Oct. 25-28, 2023.
Work Experience
-
Learning Research Assistant at Phison
(2023.07 - Present)
As a Learning Research Assistant at Phison, I am involved in some LLM system optimization project and aiDAPTIV+ solution, which allows customers to use SSDs as a cost-effective cache for model training. In my role, I have made several contributions:
1. Improved the original implementation of CPU buffer usage, reducing DRAM usage by approximately 60% through mitigating fragmentation issues. The results is depends on different model configurations.
2. Implemented a low-bit optimizer, resulting in a 50% reduction in data transfer time between SSD and DRAM.
3. Lead a team in replacing the original filesystem-based SSD I/O implementation with a manual management system for SSD data location during swap-out operations. This enhancement paves the way for future Compute-in-Storage feature implementations.
-
Part-time Software Engineer at LooPick
(2022.10 - 2023.06)
At LooPick, I worked on developing a web management interface and a LINE rental system for a city-wide reusable cup program in beverage shops. This role enhanced my skills in front-end and back-end development.
-
Summer Intern at Institute of Information Science, Academia Sinica
(2022.07 - 2022.09)
During my summer internship at Academia Sinica, I participated in research projects related to computer science and artificial intelligence. I assisted senior researchers, conducted literature reviews, and contributed to analysis for ongoing studies.
-
Learning Research Assistant at Institute of Information Science, Academia Sinica
(2021.07 - 2023.06)
As a Learning Research Assistant at Academia Sinica, I contributed to long-term computer science research projects. My primary focus was on improving the performance of Skyrmion Racetrack Memory (SK-RM) for deep learning applications. The results of this research were published at ISLPED 2023.
Educations
-
PhD at Department of Computer Science, National Yang Ming Chiao Tung University
(2025.09 - )
-
Master at Institute of Artificial Intelligence Innovation, National Yang Ming Chiao Tung University
(2023.09 - 2025.06)
GPA 4.26/4.3
-
Bachelor at Department of Computer Science and Information Engineering, National Taipei University of Technology
(2019.09 - 2023.06)
GPA 3.77/4
Open Source Contribution
-
GPT Researcher
It is a Multi-Agent LLM workflow designed to utilize various source documents and LLMs to generate comprehensive research reports.
By addressing the issue of redundancy in the researched report (#548), Implement a flow (#713) that can temporarily store written data and retrieve relevant information when writing new sections. Improve the overall quality of the report.
Skills
- Python
- Pytorch
- C/C++
Contact Me
If you have any questions or comments, please feel free to contact me.
- Email: tomhot246@gmail.com
- GitHub: DandinPower
- LinkedIn: YongChengLiaw