Title: Achieving Human Parity Performance in Pattern Recognition and Language Understanding by Machines

Abstract:
For more than a half century, computer scientists have been attempting to train computer systems to perform human perception and cognition tasks, such as, recognize image and speech, comprehend text, translate languages, etc. But until recently those systems were plagued with stagnated accuracies that were far below human performance. In recent years, with the breakthroughs in Deep Learning, advances in the state-of-the-art performance of those systems have gained a strong momentum, thanks to the rapid increase in computing power, big data, and advances in machine learning algorithms. Today, AI breakthroughs are coming at an accelerated pace. The performance of computer systems on several perception and cognition tasks has reached human parity. For example, in 2015 Microsoft researchers achieved 96% accuracy in the ImageNet Computer Vision Challenge, which is as good as a Stanford graduate student. Less than a year later, Microsoft's speech recognition system achieved 5.1% error rate on the Switchboard dataset, which is at parity with professionals who do transcription! In January 2018, Microsoft was the first to achieve human parity in text comprehension tasks on the Stanford Question Answering Dataset. And two months later, Microsoft announced that it reached human parity in English-to-Chinese and Chinese-to-English machine translation on the news dataset.  In this talk, I will briefly describe our journey to achieving human parity on these tasks and the technologies that enabled the breakthroughs. I will also present other applications of Deep Learning, such as OCR in unconstrained environments and Advertising.

Biography:
Dr. Jianchang (JC) Mao is Corporate Vice President of Bing Ads Marketplace & Serving, Artificial Intelligence & Research division at Microsoft. He leads a global team of engineers, scientists, product managers, marketplace operators, and analysts, responsible for building technologies and products, and running multi-billion-dollar advertising marketplace that powers Bing, Yahoo!, AOL, and other syndication partners.
 
Prior to joining Microsoft, Mao was Vice President and Head of Advertising Sciences at Yahoo! Labs, overseeing the R&D of advertising technologies and products. He was also the science and engineering director responsible for the development of backend technologies for several Yahoo! social search products, including Yahoo! Answers. At Yahoo!, Mao received the Leadership Superstar Award in 2010, and received a Superstar Team Award in 2008. Prior to joining Yahoo!, Mao was director of emerging technologies and principal architect at Verity Inc., a leader in Enterprise Search (acquired by Autonomy and then acquired by HP), from 2000 to 2004. Mao began his career as a research staff member at the IBM Almaden Research Center from 1994 to 2000, after receiving his PhD degree in computer science from Michigan State University in 1994.
 
Mao's research interests include AI, machine learning, data mining, information retrieval, computational advertising, pattern recognition, and image processing. He has published more than 50 papers in journals, book chapters, and conferences, and holds 29 U.S. patents.  Mao received an Honorable Mention Award in ACM KDD Cup 2002 (Task 1: Information Extraction from Biomedical Articles), an IEEE Transactions on Neural Networks Outstanding Paper Award in 1996 (for his 1995 paper), and an Honorable Mention Award from the International Pattern Recognition Society in 1993.  Mao is a Fellow of IEEE.