CAREER: Modeling Group Dynamics in Multi-agent Systems

点击翻译按钮获取中文摘要

Predicting the behavior of an individual amidst a large number of other people, such as a player in a multiplayer online game or a pedestrian in an urban scene, poses a difficult challenge. This is because an agent's actions are strongly influenced by others in its environment, even though they do not formally share goals or plans. The majority of prior work on multi-agent behavior recognition has either ignored the influence of an agent's social context or has focused on tightly-coupled agent teams. In contrast, the research supported under this award is making fundamental progress in activity recognition and synthesis for humans in large, loosely-coupled groups by creating computational models of group identity that inform agent action selection. Models are being evaluated on problems such as recognizing group membership in video footage, analyzing player strategies in multiplayer online games, and other social simulations. This research aims to make strong progress towards a better understanding of group dynamics and processes in human psychology. Educational initiatives of this award are introducing students to the exciting area of socially-interactive agent systems, using agent-based simulations to disseminate information about group bias and stereotypes (raising awareness of these issues has been proven to reduce the impact of bias on decision-making), continued curriculum development on multi-agent systems, the creation of public datasets of multi-agent social interactions, and the expansion of outreach/mentoring programs for women in computer science.This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5).