Learning and Harnessing the Power of Human Computation *note conference room change

When:
Tuesday, February 28, 2017, 11:00 am - 12:00 pm PDTiCal
Where:
6th large floor conference room
This event is open to the public.
Type:
AI Seminar - Interview talk
Speaker:
Yang Liu
Description:

Human computation advances the meet of human intelligence with machine intelligence. For example, crowdsourcing provides a cheap way for collecting large scale data. However due to the salient features of a human computation system, including no monitoring of working process, lack of ground-truth, and incentive misalignment etc, such a system suffers from severe quality control issues, and makes the online hiring process hard. Realizing the issues, researchers have proposed different methods for remediation. Nonetheless we observe a clear disconnect where Machine Learning techniques have been largely ignored for helping resolve above issues.

In this talk I will focus on designing learning systems that can help collect better quality data from human computation. The first is a double exploration aided online learning algorithm that helps select better quality workers. Going further down to the root of the problem, I then seek a more organic solution that can incentivize workers to contribute better quality data. I demonstrate that machine learning techniques provide incentives more efficiently, compared to other existing non-ML methods where redundant assignments are often required. Lastly I show that using a particular bandit learning framework, we can build a robust reputation system for incentivizing high quality data contribution harnessing the power of long term incentives, in contrast to classical one shot payment-only system. 

bio:

Yang Liu is currently a postdoctoral fellow at Harvard University. He obtained his Ph.D degree from the Department of EE:Systems, University of Michigan Ann Arbor in 2015. Before came to Ann Arbor, he got a Bachelor degree from Shanghai Jiao Tong University, China in 2010. Then he obtained his Master of Science in EE:Systems and Mathematics in 2012 and 2014 respectively, both from University of Michigan. He is author and coauthor of several technical papers in top journals and conferences of IEEE/ACM/USENIX/AAAI. His main research efforts include developing learning theory and approaches toward acquiring and processing large scale and noisy data. He is also interested in cybersecurity and decision making problems in networks. He received the best application paper award from DSAA 2014 and the best poster award from Michigan Enginnering Graduate Symposium 2011.

« Return to Upcoming Events