直播 | 京东探索研究院院长陶大程：深度学习的基础与可信AI

中科院AI科学前沿论坛

9月12日

10:00-12:00

第八讲

 Foundations of Deep Learning and Trustworthy AI

Prof Dacheng Tao

President of the JD Explore Academy 

 Vice President of JD.com

Prof Dacheng Tao is inaugural president of the JD Explore Academy and a senior vice president of JD.com. He is also an advisor and chief scientist of the digital science institute in the University of Sydney, a distinguished visiting professor to Tsinghua University, a grand master professor of University of Science and Technology of China.

He is mainly engaged in research of trustworthy artificial intelligence, and has published more than 200 papers in leading journals and top conferences. His papers have been cited more than 60,000 times. He has an h-index: 132, and received the best paper awards and test of time awards from top conferences and journals many times.

He received a 2015 Australian Eureka Prize and the 2015 UTS Vice Chancellor's medal, the 2018 IEEE ICDM research contribution award, a 2020 Australian Eureka Prize and the University of Sydney Vice Chancellor's award for research excellence, and the 2021 IEEE Computer Society Edward J McClusky Technical Achievement Award. He is a Fellow of the IEEE, AAAS and ACM, a foreign member of the Academia European, a Fellow of the Royal Society of NSW, and a fellow of the Australian Academy of Science.

ABSTRACT

Deep learning can learn very complex functions by adding more layers and more neurons to the network structure. In the past ten years, it has made remarkable achievements in various fields. Although deep learning has achieved great success in practice, there is still a lack of explanations at theoretical level. Previous works have theoretically and empirically shown that deep neural networks have sufficient ability to memorize training data with random labels. But why deep neural networks can still generalize to new data well when the number of parameters is significantly greater than the training sample size? There is still no clear explanation for this problem. To understand the generalization ability in deep learning, the academic community has proposed many complexity measures to evaluate the capacity of neural networks, such as VC dimension, spectral norm, and sharpness. However, these metrics are usually related to network size and will fail when directly used in analyzing very large models. Understanding why deep learning is capable of setting off the third wave of artificial intelligence is essential to better understand the latest advances in artificial intelligence. Today, I will share our recent research progress, proposals and insights to explain why deep learning is so successful.

时 间

2021年9月12日  周日

10：00--12：00

观看方式

腾讯会议

ID：687 734 760

会议链接：

https://meeting.tencent.com/dm/JjbAlJfuQzCm?rs=25

或扫描入会，会议二维码：