Probabilistic Graphical Models for Aggregating Crowdsourced Data
Event details
| Date | 16.06.2014 |
| Hour | 09:00 |
| Speaker | Dit-Yan YEUNG, The Hong Kong University of Science and Technology, Hong Kong |
| Location | |
| Category | Conferences - Seminars |
Crowdsourcing is a problem-solving approach which takes advantage of the wisdom of the crowd by enlisting a crowd of contributors to get a task done. Some major challenges in crowdsourcing include how to evaluate the contributors and their contributions in the absence of ground-truth information and how to aggregate the contributions of multiple contributors with the goal of outperforming any single contributor. We consider two novel crowdsourcing tasks in this talk. The first one is a video annotation task which seeks to track an object of interest as it moves around in a video. Unlike the relatively simple classification and regression tasks considered by most existing crowdsourcing methods, video annotation is significantly more complex because it involves structured time series data. The second task is peer grading which is especially crucial to the grading of open-ended assignments in massive open online courses (MOOCs). This task is unique in that the peer graders (contributors) are themselves students whose submissions are graded by their peers. We propose probabilistic graphical models for these two crowdsourcing tasks. While the models are different due to the nature of the tasks, a subtle similarity is that learning the reliability of each contributor from data plays a central role in both machine learning models.
Links
Practical information
- General public
- Free
Organizer
- Boi Faltings
Contact
- Sylvie Thomet