At a time when AI was not a buzzword, and even a little unknown, Zhu Jun began to devote himself to this area. Without a background, he chose this completely out of interest and it turned out that Zhu was not let down.
For a long time, Zhu Jun has been dedicated to the study of Bayesian AI and made great progress in it. The Bayesian regularization proposed by Zhu Jun was considered to have broken the classic Bayesian “a priori equal probability” framework, which has a history of over 250 years, provided the third dimension degree of freedom for Bayesian inferences, and brought brand new orientation in application. Under the regularized Bayesian framework, Zhu’s maximum margin learning theory and efficient algorithm based on Bayesian models have combined the two major directions which were separated for over 20 years.
As “the most distinguished PhD student Zhang Bo, a Tsinghua University professor in the department of computer science and academician of CAS (Chinese Academy of Sciences), has ever coached,” Zhu Jun was invited by Zhang back to Tsinghua for teaching in 2011 and was in charge of cultivating the next generation of PhD students. Shortly after his return, Zhu demonstrated his ability in AI research and talent cultivation and became a key figure in China’s AI circle.
In 2017, a Tsinghua University team led by Zhu Jun published an abacus programming probability library, which further integrated the Bayesian method with deep learning and realized multi-machine and multi-GPU efficient computing.
By now there has been a lot of open framework which can support deep learning in developing and prototype designing. But there are seldom any platforms that can support Bayesian deep learning. According to his planning, Zhu hopes to build a platform called “Abacus." On the one hand, it can support deep learning. On the other hand, it can also support Bayesian inferences. More importantly, this will be an integrated platform between deep learning and Bayes and promote related research and engineering practices.