Abstract: This talk will review recent accomplishments from joint work being carried out by CSU, The University of Florida and Brandeis University.  The work highlights the importance of agents, i..e artificially intelligent assistants, using all available senses to share a common view of our surroundings and shared tasks. This most particularly means paying  attention not just what  we are saying, but how we are saying it as expressed through tone and facial expressions, what we are looking at and finally how we may be using our bodies to communicate.  Recent findings in both human studies and machine-human interaction experiments will show that when solving physical problems in the world gestures play a critical role.  Specifically, the time required by people to jointly build a structure out of blocks under the following conditions is roughly the same.  Condition 1 has each person able to see the other but they cannon speak.  Condition 2 has each person able to hear the other but they cannot use gestures.   These findings underscore the importance of agents doing more than just listening and speaking.  

Bio: Professor Beveridge works on Computer Vision, emphasizing problems relating to recognition and understanding. As computer vision plays an ever more important role in the development of agents, it is clear that blind agents will fall by the wayside as everyday helpers; replaced by agents able to recognize us on sight, respond accordingly to our expressions and gestures, and most important discuss a shared understanding of the objects both they and we can see. Professor Beveridge's current work on how computers may better communicate with people through sight, gesture and language draws upon past work in video understanding, human face recognition and object recognition. Algorithm evaluation is one key part of this past work, including face recognition in video, and older legacy work supplying open source tools such as the CSU Face Identification Evaluation System. This legacy system, release in 2001, has been downloaded over 24,000 times since its introduction in 2001. As a teaching tool, in 2009 an open source live video face recogntion system, FaceL, was released to the web that incorporates new innovations in the development of correlation filters. Professor Beveridge's other interests include high dimensional data analysis, optimal matching of geometric features, genetic algorithms and the use of reconfigurable embedded hardware. Professor Beveridge maintains a set of web tools for deploying public-facing course websites.