Computational Models for Multimedia Analysis

Sometimes, human decision-making processes are not as rational as it seems. Each person has his/her own experience, knowledge, and preference on things and events surrounding us. This idiosyncratic behavior is hard to quantify and analyze and thus exists as an important obstacle to the complete understanding of human decision process and eventually implementation of more realistic artificial intelligence.

Understanding high-level concepts, e.g., interestingness, aesthetic beauty, attractiveness, atypicality, etc., and computational models to automatically estimate such concepts from complex, structured information are the essential first steps to reach the goal of understanding the irrational human behavior and decision making processes. To this end, we are investigating ways to measure and estimate these concepts from multimedia data.