Oscar, Li

Oscar, Li 李勝維

ML Algo. Engineer @ Google

About Me

Hi! I’m Oscar, a machine learning algorithm engineer at Google. I specialize in vision + X multimodal learning and efficient Gen-AI.

My track record includes working with industry leaders like Google, Intel, and TSMC. I also have published research papers in top-tier conferences such as ECCV. I am always seeking opportunities to translate cutting-edge technology into significant value.

Download my Resume
Interests
  • Vision + X Multimodal Learning
  • Large-Scale LLM Training
  • Efficient Diffusion Model & LLM
Education

  • M.S. in Computer Science, 2024

    National Taiwan University

  • B.S. in Computer Science, 2022

    National Chiao Tung University

Experience

 
 
 
 
 
Google
Machine Learning Algorithm Engineer
Google
May 2025 – Present
Created ML algorithms, drove hardware- and silicon-aware optimizations influencing future SoC designs, and generated patentable intellectual property.
 
 
 
 
 
Google
Machine Learning Intern
Google
June 2024 – September 2024
Developed Generative AI for image editing using text prompts and designed an efficient generative model architecture achieving 25x faster inference.
 
 
 
 
 
Intel
Software Engineer Intern
Intel
July 2023 – May 2024
Contributed to the internal LLM copilot for unit-test generation, resulting in a 20% improvement in code coverage.
 
 
 
 
 
NARLabs 國家實驗研究院
LLM Engineer, TAIDE LLM
NARLabs 國家實驗研究院
March 2023 – September 2023
Developed the TAIDE LLM for Taiwanese culture. Involved in preference data generation, pre-training, instruction tuning, and safety alignment.
 
 
 
 
 
Tomofun
AI Engineering Intern
Tomofun
March 2023 – June 2023
Focused on the efficient tuning of a vision-language model for pet monitoring, improving the model’s out-of-distribution factuality by 15%.
 
 
 
 
 
TSMC and ACM Lab, NCTU
Research Assistant
TSMC and ACM Lab, NCTU
January 2021 – January 2022
Engaged in efficient meta-learning (AI Learning and Digital Twin in HVAC Control) for sensor forecasting in HVAC control, significantly reducing training data requirements by 97.5%.

Honors

The Phi Tau Phi Scholastic Honor Society
Honorary member of the Phi Tau Phi Scholastic Honor Society
The Phi Tau Phi Scholastic Honor Society Jun 2024
Honored for exceptional academic performance, ranking among the top 1% of students.
NCTU & The Zhu-Ming Academic Foundation
The Yin Zhi Tong Memorial Scholarship
NCTU & The Zhu-Ming Academic Foundation Aug 2022
Awarded for outstanding performance in the AI Learning and Digital Twin in HVAC Control project.

Publications

Sheng-Wei Li, Zi-Xiang Wei, Wei-Jie Chen, Yi-Hsin Yu, Chih-Yuan Yang, Jane Yung-Jen Hsu (2024). [ECCV 2024] SA-DVAE: Improving Zero-Shot Skeleton-Based Action Recognition by Disentangled Variational Autoencoders. In ECCV 2024.

Cite Code

Projects

*
All Large Language Model Competition
Taiwan-ELM

Taiwan-ELM

Efficient LLM for Taiwan with open weights, datasets, and checkpoints and affordable sizes (270M/1.1B).

Huggingface Datasets
TAIDE - Trustworthy AI Dialog Engine

TAIDE - Trustworthy AI Dialog Engine

The TAIDE project developed a trustworthy LLM tailored for Taiwanese culture. Outperforming GPT3.5-Turbo on 14 task evaluations.

Huggingface Project Main Page News
哈哈狗, an Emoji Language Model Bot

哈哈狗, an Emoji Language Model Bot

The bot inserts emojis in the given paragraph using trained language model.

Code
Listen Watch and Read: Multimodal Information Authenticity Detection and Source Tracking

Listen Watch and Read: Multimodal Information Authenticity Detection and Source Tracking

Proposed dataset and system to detect misinformation and AI-generated contents in text, video, and audio.

PDF
ScanNet200 Long-tail Point Cloud Semantic Segmentation Challenge 2022

ScanNet200 Long-tail Point Cloud Semantic Segmentation Challenge 2022

Combined attention and U-Net for long-tail point cloud semantic segmentation. Voted 1st place in the Challenge for best novelty and poster presentation.

Code Challenge
AI Learning and Digital Twin in HVAC Control

AI Learning and Digital Twin in HVAC Control

Digital twin and AI for HVAC control. The research won the Gold Award of the Intelligent Living Space Design Competition in 2022 for its effectiveness, efficiency and novelty.

News Competition