יום חמישי, 13.12.2012, 13:30
חדר 1061, בניין מאייר, הפקולטה להנדסת חשמל
For decades, our culture has been fascinated with the concept of the “Star Trek computer”. An all-knowing entity which is available to query about almost anything relevant to the world around us, organize our daily lives, remind us when we are doing the wrong thing, help us when we are lost. Several key features of this type of system are: (1) We can interact with it via natural language and it will understand our words (speech2text), (2) it will understand what we want to know (natural language processing), (3) it will recognize objects and entities about which we are querying (object recognition) and (4) it will know about entities in the real-world about which we may be querying (knowledge representation). The technical fields required for this type of system have made great individual progress over the past 20-30 years, but their capabilities are still frustratingly limited. Over the same time, in AI there has also been much progress in uncertain reasoning with graphical models, multi-view learning, multi-task learning and relational learning, which aim to combine many related sources of information together to improve classification ability. In this talk, I argue that by combining different modalities and different classifiers, we may enable a qualitative shift in our perception ability and may bring the Star Trek computer within reach for personal use.
Denver Dash is a Research Scientist at the Intel Science and Technology Center on Embedded Computing (ISTC-EC) and adjunct faculty in Robotics at Carnegie Mellon University. Denver has been at Intel since 2003 when he was hired by Gary Bradski as a research scientist and technical director of Intel’s Open Probabilistic Network Library (OpenPNL) in Santa Clara, CA. In 2006 he moved to the Intel Research Lablet at Carnegie Mellon, and in 2012 he was a founding member of the ISTC-EC. In general, his research interests lie in the intersection of machine learning, probabilistic graphical models, causality and first-order logic. While at Intel he has applied these tools to learning for silicon manufacturing, disease outbreak detection, distributed detection and inference of network intrusions, multimodal perception and causal perception. He frequently sits on the program committees of most top conferences in artificial intelligence and machine learning, most recently serving on the senior program committees for IJCAI-2013 and UAI-2012. For a full list of publications, see http://www.pittsburgh.intel-research.net/~dhdash/pubs/pubs.html.