About
Shawn is a highly experienced technology professional with a passion for engineering and algorithms. Prior to joining a stealth startup in 2022, he was a founding member of Kuaishou Technology (SEHK:1024) AI Platform, where he directed its Engineering with emphasis on Personalization Infrastructure and successfully led multiple core teams to drive innovative solutions. With his expertise, he developed the first GPU-based large scale advertising recommendation system at Kuaishou, which generated a revenue of 6 billion USD annually. In 2021, he created the world's largest scale recommender system (ACM SIGKDD 2022). He also served as a Senior Staff Research Scientist at Kuaishou Seattle AI Lab, where he was instrumental in inventing and developing cutting-edge systems utilizing the latest technologies, including large scale storage engines, high performance deep learning infrastructure, and model compression frameworks. His strong leadership and expertise in the field make him a valuable asset to any team.
In addition to his extensive experience in the technology industry, Shawn holds a PhD degree in Computer Science and Artificial Intelligence from the University of Rochester. During his academic pursuits, he made notable contributions to the field of distributed machine learning, including publishing important results on the theoretical justification of asynchronous SGD (NeurIPS 2015 spotlight ) and the first decentralized SGD with linear speedup (NeurIPS 2017 oral ). He received the 30 New Generation Digital Economy Talents (30 位新生代数字经济人才) award from World Internet Conference and Big Data Digest in 2019.
Shawn's expertise in technology is rooted in a lifelong passion for programming that began even before he started preschool. This comprehensive engineering knowledge, combined with a deep understanding of algorithms, has enabled him to guide teams to successful outcomes on even the most challenging real-world problems. Throughout his career, he has consistently demonstrated his ability to apply his technical know-how and leadership skills to drive impactful results and make a significant impact in his field.
Selected Projects
-
PERSIA
Kuaishou's advanced GPU-based large scale learning system designed for
ad recommendation and CTR prediction tasks. Launched in 2018 by Shawn,
PERSIA has been leading the charge in the field of recommendation
systems, and was open-sourced in 2021. With the ability to support
models with up to 100 trillion parameters, PERSIA is the fastest public
recommendation model training framework available. Built using Rust for
high performance computing and communication, PERSIA is a testament to
Kuaishou's commitment to advancing the state of the art in
recommendation systems.
- Training Deep Learning-based recommender models of 100 trillion parameters over Google Cloud.
- PERSIA, the largest recommended training system in the history of open source by far.
- Story: 640x Faster GPU Based Learning System for Ad Recommendation.
- Story: Innovation, Balance, and Big Picture: The Speed of Kwai Commercialization.
(collaborate with DS3 Lab) -
Bagua
Kuaishou's deep learning training acceleration framework designed
to tackle the challenge of large scale training tasks. Bagua offers a
comprehensive solution to speed up the training process, including data
loader optimization, advanced distributed training algorithms, network
communication optimization, and more. Developed to solve the training
bottleneck at Kuaishou Technology, where more than a million videos are
uploaded every hour, Bagua has been instrumental in maintaining the
company's position at the forefront of innovation.
(collaborate with DS3 Lab)
- Hammer The automatic deep learning model compression tool developed by Kuaishou Technology. With Hammer, reducing the size of large models while maintaining their accuracy has never been easier. This innovative tool has already made a big impact at Kuaishou, saving thousands of GPU cards and enabling the successful deployment of hundreds of complex models. (Hammer also helped the TAMU-KWAI team won the 2nd prize in IEEE Low-Power Computer Vision Challenge).
-
DPSGD/ADPSGD Algorithms
The DSPGD/ADPSGD algorithms are revolutionary decentralized training
solutions for artificial intelligence, offering unparalleled speed and
accuracy. Unlike traditional training algorithms, these decentralized
algorithms are optimized for cloud computing environments where network
conditions and machine performance can vary. As demonstrated by IBM,
the DSPGD/ADPSGD algorithms can drastically reduce training times for
speech recognition AI, from a week
to just 11 hours, while also delivering a 10x improvement in
performance.
(collaborate with IBM Thomas J. Watson Research Center )
-
DouZero
A strong game AI for DouDizhu
. The corresponding research paper was accepted by ICML 2021.
(collaborate with DATA Lab )
- cproxy A little handy tool Shawn created to apply proxy transparently on individual processes using cgroups.
Academic Professional Activities
- Senior Program Committee of AAAI
- Program Committee of ICML, NeurIPS , ICLR, AISTATS , AAAI, and ScaDL
-
Journal Reviewer of
- Journal of Machine Learning Research (JMLR)
- Machine Learning
- IEEE Transactions on Pattern Analysis and Machine Intelligence
- IEEE Transactions on Information Theory
- IEEE Transactions on Network Science and Engineering
- IEEE Transactions on Neural Networks and Learning Systems
- IEEE Transactions on Knowledge and Data Engineering
- IEEE Transactions on Signal Processing
- IEEE Internet of Things Journal
- Data Mining and Knowledge Discovery
- BIT Numerical Mathematics
- Computational Optimization and Applications
- Optimization Methods and Software
- European Journal of Operational Research
- Journal of Parallel and Distributed Computing
- Pattern Recognition
- Neural Networks
- Parallel Computing
- Neurocomputing
- Measurement
- International Journal of Electrical Power & Energy Systems
- Journal of Optimization Theory and Applications