Nice link, Aaron, That's what I was talking about with 3D-reconstruction - it involves also reconstruction of the light sources - in GI. You don't need millions of dumb samples of "all possible cases of ..." like the brute force (dumb) machine learning, the problem must be approached right with finding the appropriate correlations, then there is not a combinatorial explosion.
Different human activities are encoded in the possible motions of the human body and their possible interactions with other bodies (in classical physics sense), i.e. initially those bodies must be identified, and that must be evaluated in as lower resolution, as general the activity is. The symbols are essentially low resolution representations, they are always fewer than the raw data. The activities obviously encoded in the trajectories of those bodies (that's something physics is dealing with for centuries), and when generalized in low resolution, the trajectories would converge to the cases of carrying, throwing, kicking etc., depending on some rough/global specifics like is it accelerating, is it by a hand, an arm, head, are the arms moving in parallel etc. etc. -- ** Todor "Tosh" Arnaudov ** * -- Twenkid Research:* http://research.twenkid.com -- *Self-Improving General Intelligence Conference*: http://artificial-mind.blogspot.com/2012/07/news-sigi-2012-1-first-sigi-agi.html *-- Todor Arnaudov's Researches Blog**: *http://artificial-mind.blogspot.com > - *From:* Aaron Hosford <[email protected]> > > > - *To:* [email protected] > > > - *Subject:* Simulation for Perception, Symbols for Understanding > > > - *Date:* Tue, 30 Oct 2012 09:30:55 -0500 > > http://www.rec.ri.cmu.edu/about/news/11_01_minds.php > > Recognizing and predicting human activity in video footage is a difficult > problem. People do not all perform the same action in the same way. > Different actions may look very similar on video. And videos of the same > action can vary wildly in appearance due to lighting, perspective, > background, the individuals involved, and more. > To minimize the effects of these variations, Carnegie Mellon's Mind’s Eye > software will generate 3D models of the human activities and match these > models to the person’s motion in the video. It will compare the video > motion to actions it’s already been trained to recognize (such as walk, > jump, and stand) and identify patterns of actions (such as pick up and > carry). The software examines these patterns to infer what the person in > the video is doing. It also makes predictions about what is likely to > happen next and can guess at activities that might be obscured or occur > off-camera. > This project's approach is to use 3D simulation to detect and classify > behavior, and then generate symbolic information about the events that were > observed. I'm encouraged to see someone doing work on this stage of > cognition, as I see perception as the "missing link" that's stopping AGI > from developing. > > I wonder, will a certain naysayer feel vindicated that someone else sees > simulation as vital to intelligence (and is using it to solve precisely the > problems he says it's needed to solve), or will he be annoyed that the > ultimate form the information takes is symbolic, which is compatible with > semantic nets or any number of other existing AGI approaches? ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968 Powered by Listbox: http://www.listbox.com
