Mu Li

Career

Li earned a PhD in computer science at Carnegie Mellon University, where one of Li's advisors was Alex Smola, according to a 2017 AWS profile. The same profile records that Li had served as a principal research architect at Baidu's Institute of Deep Learning and as CTO of the AI start-up Marianas Labs before joining AWS as a principal scientist for machine learning.³

At AWS, Li led a team working mainly on the Apache MXNet framework, with a focus on making deep learning easier to use and to run at scale across many GPUs and machines. In 2023 Li and Smola founded Boson AI, which builds the Higgs multimodal models for voice and audio, and the company's team page lists Li as Co-Founder & CEO / CTO.³²¹

Research and open-source systems

Li's research has centred on large-scale machine learning and the co-design of distributed systems and learning algorithms. At OSDI 2014, Li presented "Scaling Distributed Machine Learning with the Parameter Server", written with colleagues from Carnegie Mellon, Baidu and Google, which describes a framework where server nodes hold shared parameters while workers process distributed data.³⁸

Li co-authored the 2015 MXNet paper describing a flexible library for distributed machine learning across heterogeneous hardware. Later papers with Li as a co-author include "Bag of Tricks for Image Classification with Convolutional Neural Networks" (2018) and the 2019 paper introducing the GluonCV and GluonNLP toolkits for computer vision and natural language processing.⁴⁹¹⁰

Teaching and writing

Li is one of the four main authors of Dive into Deep Learning, with Aston Zhang, Zachary C. Lipton and Alexander J. Smola. The interactive book pairs text with runnable code in several frameworks, and its site says it is forthcoming from Cambridge University Press. The source repository releases the text under CC BY-SA 4.0.⁵⁶

Work and contributions

• Dive into Deep Learning Published research

  Co-author of the open-source book, listed among its main authors; the repository releases the text under CC BY-SA 4.0 and the sample code under a modified MIT license.⁵⁶

• Apache MXNet Open-source code

  Co-author of "MXNet: A Flexible and Efficient Machine Learning Library for Heterogeneous Distributed Systems" (2015) and, per a 2017 AWS post, leader of the AWS team working primarily on the framework.⁴³¹¹

Timeline

1. Aug 7, 2014Presented "Scaling Distributed Machine Learning with the Parameter Server" at OSDI 2014.⁸
2. Dec 2015Co-authored the paper introducing the MXNet machine learning library.⁴
3. May 2017Profiled by AWS as a principal scientist leading a team working on Apache MXNet.³
4. Dec 2018Co-authored "Bag of Tricks for Image Classification with Convolutional Neural Networks".⁹
5. Jul 2019Co-authored the paper introducing the GluonCV and GluonNLP deep learning toolkits.¹⁰
6. 2023Co-founded Boson AI with Alex Smola.²

Other documented roles

Roles the sources document that are not confirmed as current.

• Principal scientist for machine learning, leading a team working primarily on Apache MXNet, Amazon Web Services³⁵
• CTO (before joining AWS), Marianas Labs³
• Principal research architect, Baidu Institute of Deep Learning³

Where they work

• Boson AI 2 people in the wiki

Related people

• Alex SmolaAlso at Boson AI
• Sylvain GuggerML frameworks and libraries · Training systems
• Luca AntigaML frameworks and libraries · Training systems
• Pieter NoordhuisTraining systems · ML frameworks and libraries
• Martín AbadiML frameworks and libraries · Training systems
• Andreas C. MüllerML frameworks and libraries · Teaching and writing
• Edward YangML frameworks and libraries · Training systems
• Frédéric BastienML frameworks and libraries · Training systems
• Jane XuML frameworks and libraries · Training systems
• Alban DesmaisonML frameworks and libraries · Training systems