'Efficient Plan Recognition for Dynamic Multi-agent Teams'
Gita Sukthankar (University of Central Florida), Katia Sycara (Carnegie Mellon University)
Gita kicked off the New Investigators in AI session this Friday morning by addressing the problem of multi-agent plan recognition. It wasn't an area I knew anything about, but to my mind team dynamics is generally an interesting topic.
She spoke about multi-agent plan recognition, using spatial, temporal and coordination data for understanding dynamic team plans. Plan recognition is understood through observations, prior knowledge, and closed world observation, but there are problems with understanding dynamic membership, especially when there is team change or subteams. She mentioned that sub-team creation and destruction are important cues and spoke about ways to leverage temporal, spatial and coordination cues for understanding dynamic team membership.
She took advantage of her phone ringing partway through her talk to prompt for questions, which was skilfully managed, and won my respect for her presentation skills.
----
'Simplifying Sketch Recognition UI Development
'
Tracy Hammond (Texas A&M University)
Tracy gave a lovely and engaging presentation. As she pointed out sketching is engaging, creative and involves active learning. Sadly sketching is generally dropped from the curriculum as hard to test and correct. Sketch-recognition systems are a way to bridge this gap - by looking at the identification of hand-drawn shapes.
Tracy gave a really interesting demo of sketching to create moving demos using LADDER Domain builder (LADDER is a sketch recognition language). Such systems are quick and fun ways to create natural design animations, give immediate feedback, reduce teacher time and enhance teacher feedback. Systems like this are very hard to build. It is impractical for instructors to build such systems, so her work has been looking at ways to simplify this process.
I had never really focussed on how hard stroke and shape recognition could be, especially for arcs, but the demo and videos she gave were really impressive. She mentioned her student's poster on recognising scribble types (for filling vs. deleting scribbles), research on natural shape descriptions, and showed videos using sketch recognition.
The GUILD system has been created to generate user interfaces using LADDER descriptions and Tracy went on to demonstrate how it could be used to create a tic-tac-toe sketch recognition board game. So far 30 systems have been built using this system across many domains, and has been used for student prototyping.
I really enjoyed this talk and it really helped me understand both the problems with, and potential uses for, such systems and I hope to read more about this on the Sketch Recognition Lab site later.
----
Johanna Hunt
Official GHC 2007 Blogger
You may comment on this blog by visiting the GHC Forum
Comments