For more information, please visit my group's website: Computational Behavior Group.
Multi-agent Learning and Human Behavior
Learning and adaptation are crucial hallmarks of human behavior. While machines cannot replicate human intelligence, learning algorithms can potentially model and simulate certain human behaviors with more realism than traditional approaches such as game theory or psychological factors alone. For example, no-regret algorithms have been shown to realistically model human decision making in social settings. Through human subject experiments both online and in lab settings, we are exploring ways of using machine learning techniques to leverage the growing size of available social behavior data. These datasets enable us to explore a myriad of phenomena, ranging from economic network interactions to inference of moral values based on experiment/game behavior.
Collaborators: Rajiv Maheswaran, Jesse Graham, Ravi Iyer, USC; Leslie Kaelbling, MIT.
Pattern Recognition and Anomaly Detection in Spatio-Temporal Data
Movement data can be combined with geospatial information and transformed into probabilistic graphical models that represent both social and temporal relationships between objects in the observed area. We then apply machine-learning techniques to cluster patterns in these graphical models to assist human users in performing strategic level analysis such as behavior prediction and anomaly detection. As an example, we are taking a subset of Twitter data that includes geo-location tags, and analyze behavior patterns of users in major cities across the world.
Collaborators: Rajiv Maheswaran, Craig Knoblock, Pedro Szekely, USC
Cognitive AI
We have been working on developing a system we call "Jean".
Jean is an architecture that models developmental learning in an autonomous agent platform. Jean acts
and learns in simulated 3D environemnts, building up a richer understanding of the world her over time as she interacts with her environment.
Image schematic representations, and learning algorithms based on these representations, form the core of Jean's capabilities.
Collaborators: Paul Cohen, Clayton T. Morrison, USC ISI; Robert St. Amant, NCSU.
Mobile Ad-Hoc Networks
We have applied reinforcement learning techniques to problems of routing and movement in mobile ad-hoc networks, and showed that nodes that learn to cooperate to accomplish simple targeting and relaying tasks.
Collaborators: Tracey Ho, Caltech; Leslie Kaelbling, MIT.
Interactive Games
Games and the Internet offer a new means of conducting AI research. Millions of humans are available online, as long as they are entertained, and could serve as teachers for learning agents. Simulation technologies in games have developed to a point where games can serve as a realistic test-bed for AI techniques. We are developing WubbleWorld, an online game in which children can teach a cute learning agent to do varioius tasks in a 3D environment. We are also developing ISIS, a 3D real-time strategy game that serves as a military battlefield simulation environment. Finally, we are working with USC's GamePipe program to develop a Cosmopolis-based game designed to investigate models of human decision-making.
Collaborators: Wesley Kerr, Daniel Hewlett, U. Arizona; Mike Zyda, Marc Spraragen, Balki Ranganathan, USC; Travis Ho, NUS.
K12 STEM Education
Technology, and machine learning in particular, offer the promise of improving K12 education. First, integrated data can provide a more comprehensive view of student performance, and ML techniques can predict individual performance and recommend interventions. This kind of data-driven differentiated instruction can tailor learning experiences for individual needs and preferences, hoping leading to improved outcomes.
Second, intelligent systems can recommend mentors (both teachers and students) to each other in online social networks, as well as recommending content from repositories, blogs, and discussion forums.
Collaborators: Jihie Kim, Pedro Szekely, John Callahan, Winnie Callahan, USC ISI; Kim Lightle, Ohio State.
