Title : OneFlow: A Novel Deep Learning Framework Born for Large Scale Distributed Training 

Jinhui Yuan

Abstract : 
Recently, OneFlow was open sourced on GitHub. Given that we already have TensorFlow and PyTorch, people may ask such problems: is OneFlow yet another deep learning framework? What is the add-on value of OneFlow compared to existing deep learning frameworks? What is the core idea behind OneFlow? In what cases, should developers use OneFlow instead of others? People who has interest can find some useful introductions here. In this tutorial, I will try to answer the above questions and introduce how to use OneFlow to easily solve several problems which are extremely challenging to existing deep learning frameworks.

Biography :
Jinhui Yuan, Founder and CEO of OneFlow Inc, committed to developing the next generation deep learning framework OneFlow. Jinhui is the chief architect of the open-sourced Dubhe AI platform proposed by Zhejiang Lab. Jinhui got his PhD from department of computer science from Tsinghua University in 2008 (with honor). He has published dozens of papers on computer vision and multimedia retrieval in top conferences and prestigious journals. He once ranked 1st in TRECVID organized by NIST. In 2013, Jinhui joined MSRA and started to carry out some in-depth research on distributed system for large scale machine learning. In 2014, Jinhui and collaborators invented LightLDA, which was once the fastest system for training large scale topic model. Since 2015, Jinhui focused on building a new generation deep learning framework and got the special award of MSRA director (1%).