Tzu-wei Sung
Software Engineer at Google
Sunnyvale, California, United States
Summary
🤩
Rockstar🎓
Top SchoolTzu-wei Sung is a software engineer with 11 years of experience, currently working on the TPU compiler at Google after roles at Apple and research positions at National Taiwan University. He combines strong academic pedigree (NTU BS, UCSD MS with top grades) with hands-on ML engineering, contributing to prominent open-source projects like TensorFlow Addons and the CleverHans adversarial example library. His work spans low-level performance ports and API migrations, end-to-end speech systems, and robust feature engineering—evident from contributions implementing Mel-spectrogram/MFCC pipelines and migrating custom ops to tf.function. He has a track record of improving model quality and maintainability, from reducing perplexity in code-switching language models and lowering word error rates to updating TensorFlow compatibility and adding tests. Based in Sunnyvale, he brings both research rigor and production experience to compiler and ML infrastructure challenges.
11 years of coding experience
2 years of employment as a software developer
Bachelor of Science - BS, Computer Science, 4.16 / 4.30, Bachelor of Science - BS, Computer Science, 4.16 / 4.30 at National Taiwan UniversityUniversity of California, San Diego
Chinese, English
Github Skills (15)
computer-vision10
pytorch10
machine-learning10
audio-processing10
speech-recognition10
tensorflow10
python10
bert10
image-processing10
natural-language-processing9
security9
unit-test9
benchmarking8
benchmark8
sentencepiece8
Programming languages (4)
Github contributions (5)
tensorflow/addons
Mar 2019 - Feb 2021
Role in this project:
ML Engineer
ML EngineerContributions:255 reviews, 125 commits, 406 PRs in 1 year 11 months
Contributions summary:Tzu-wei's primary contribution involved migrating and testing the `distort_image_ops` to a new system. This migration included porting the custom operations, integrating the `tf.function` decorator for improved performance and maintainability, and adding comprehensive tests. The user also addressed code formatting, internal API, and copyright issues. Furthermore, they worked on migrating other image-related functionalities, such as dense and sparse image warp.
python2-xdeep-learningaddonssig
Alexander-H-Liu/End-to-end-ASR-Pytorch
Aug 2019 - Oct 2019
This is an open source project (formerly named Listen, Attend and Spell - PyTorch Implementation) for end-to-end ASR implemented with Pytorch, the well known deep learning toolkit.
Role in this project:ML Engineer
ML EngineerContributions:25 commits, 20 pushes in 2 months
Contributions summary:Tzu-wei primarily contributed to the development of audio feature extraction and text encoding modules for an end-to-end ASR system implemented in PyTorch. They added functionality for creating Mel-spectrograms, MFCCs, and delta features, enabling the use of various audio processing techniques. Furthermore, the user implemented different text encoders (Character, Subword, Word and Bert) to handle text data for the ASR model, including the ability to load pre-trained BERT models. Finally, the user added tests to validate the functionality of the new modules.
pytorchend-to-endasrdeep-learningctc
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.