[4eyes] Reminder: (Today) Research Group Talk - John Tsotsos

[Matthew Turk]

Reminder – today at 3:30pm. The HFH doors may be locked, but we’ll try to keep them open around 3:30pm.

 

Monday February 15, 2016 (President's Day*)

3:30 - 4:30pm

1132 Harold Frank Hall

 

Active Perception: 

A Necessary Component of Robotic Cognition

 

John K. Tsotsos, York University (Toronto, Canada)                 

 

A modern perspective on computer vision, robotics and artificial intelligence tells us that the active approach to perception and intelligence, as articulated in the late 1980's, is outdated. The combination of modern sensors and effective machine learning and classifier techniques seems to eliminate the need for active, interpretation-driven, visual sensor control. The supporting evidence is impressive; still, there seems to be something missing.  Keynote presentations at ICRA 2105 highlighted this: robots just can't figure things out. They cannot interpret their world in the context of their task, knowledge and surroundings. If you care about human vision and how it might inform machine vision, then it’s clear: the current paradigm does not suffice. And if you don’t care about human vision, the challenge remains to design a limited resource system that is a multi-tasker, that not only acquires and reacts to data, but does so in support of and under the direction of visual cognition (including planning, inductive generalization, recognition, and more). A passive sensing strategy, no matter how much data is collected, gives up control over the quality and specific characteristics of what is sensed and at what time and for which purpose. Passive sensing reduces or eliminates the utility of any form of predictive reasoning strategy (hypothesize-and-test, verification of inductive inferences, including Bayesian, etc.).  Even in the light of the latest accomplishments in machine perception and robotic control, it seems too early in the robot intelligence story to believe such reasoning strategies are ultimately unnecessary. This presentation will highlight the work in my lab over the past 30 years that addresses these issues and will introduce a proposal for how the missing elements may be addressed that naturally involves a new excursion into active perception.

 

Prof. John K. Tsotsos is currently the Distinguished Research Professor of Vision Science at York University, where he also holds the NSERC Tier I Canada Research Chair in Computational Vision.   He holds Adjunct appointments in the departments of Computer Science and of Ophthalmology and Vision Sciences at the University of Toronto and at the Toronto Rehabilitation Institute. He was Director of the highly respected Centre for Vision Research at York University from 2000 to 2006. His research efforts span the areas of computer vision, computational neuroscience, human vision, artificial intelligence and robotics. He is recipient of the 2006 Canadian Image Processing and Pattern Recognition Society Award for Research Excellence and Service, and of the 1st President’s Research Excellence Award by York University in 2009. He was elected Fellow of the Royal Society of Canada, Academy of Sciences, Division of Mathematics and Physical Sciences in 2010. Prior to joining York University in 2000, he was Professor and Associate Chair of Computer Science at the University of Toronto, where he began his academic career in 1980. While there, he was also appointed to the Division of Cardiology, Faculty of Medicine, and was Fellow of the Canadian Institute for Advanced Research from 1985-1995. He has published over 300 papers, 6  of which have received distinctions. His most recent research monograph "A Computational Perspective on Visual Attention" was published by The MIT Press in 2011.



*President's Day is a campus holiday

Hosts: Matthew Turk and Sven Dickinson

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.ucsb.edu/pipermail/ilab-users/attachments/20160215/0c5ff604/attachment.html>