visual processing
one of the most important problems the defense industry has yet to solve is that of Identification Friend of Foe. active methods of IFF require emitting a signal - essentially "yelling across the room: are you a good guy or a bad guy??", and waiting for an answer. if the "bogie" chooses not to answer, you still haven't confirmed anything more than he might be hard of hearing right now for any number of reasons - or he doesn't want to respond to you and "yell back" giving away his position.
consequently, the preferred method of identification would be a passive mechanism... something like the mark 1 eyeball would be awesome... and passive as long as you weren't using a flashlight. but this problem is surprisingly hard to teach to a computer. even when we teach a machine what an object looks like on all sides - we have no way for it to anticipate the orientation of that object or how much of it is exposed when one is encountered – so the computer can’t do the matching. and we can't get ahead of this.
as i think about my own object recognition capabilities - or that of a six year old, we have no problem recognizing "chair ness". when we encounter a chair obscured, or at a odd angle, we have no problem rotating it around in our head, filling in the blanks, and identifying it. this capacity eludes our programming strategies and capabilities of today.
at the day job, we’ve been trying to create a model / computer program that solves a problem that has a number of constraints - and, once we formulate those constraints, were trying to identify the maximum point - which would be our optimum. for instance, if you visualized a messed up bed - and that geometric pattern was defined by a series of data and equations, we're looking to teach a computer a smart way to find the highest point on that bed. via e.g. a heuristic, linear program, or differential equation.
since ive done little of this kind of algorithm formulation in the real world - and recall what i know from classes that were back in '99 - i find i am a wee bit rusty on some of my math. however, i am finding i am having no problem at all discussing the problem and approaches, being a sounding board, testing ideas, even making jokes - because everything we're doing has a visual allegory to it. i don't have the math handy - but all the processing and imaging can take place in the visual domain.
this caused me to think of computer graphics chipmakers nvidea, and their old slogan, "the display is the computer". and ponder the significance that visual processing is likely to make in future computer design - most especially in artificial intelligence. visual processing has been selected by nature as such a powerful tool for coping with our reality. from the series of differential equations a cheetah "intuitively" solves to intercept its prey - to even more awesome maneuvers performed by little birds twittering by us day in and day out. visual processing is hugely powerful.
i'm just saying, buy stock in nvidia.
consequently, the preferred method of identification would be a passive mechanism... something like the mark 1 eyeball would be awesome... and passive as long as you weren't using a flashlight. but this problem is surprisingly hard to teach to a computer. even when we teach a machine what an object looks like on all sides - we have no way for it to anticipate the orientation of that object or how much of it is exposed when one is encountered – so the computer can’t do the matching. and we can't get ahead of this.
as i think about my own object recognition capabilities - or that of a six year old, we have no problem recognizing "chair ness". when we encounter a chair obscured, or at a odd angle, we have no problem rotating it around in our head, filling in the blanks, and identifying it. this capacity eludes our programming strategies and capabilities of today.
at the day job, we’ve been trying to create a model / computer program that solves a problem that has a number of constraints - and, once we formulate those constraints, were trying to identify the maximum point - which would be our optimum. for instance, if you visualized a messed up bed - and that geometric pattern was defined by a series of data and equations, we're looking to teach a computer a smart way to find the highest point on that bed. via e.g. a heuristic, linear program, or differential equation.
since ive done little of this kind of algorithm formulation in the real world - and recall what i know from classes that were back in '99 - i find i am a wee bit rusty on some of my math. however, i am finding i am having no problem at all discussing the problem and approaches, being a sounding board, testing ideas, even making jokes - because everything we're doing has a visual allegory to it. i don't have the math handy - but all the processing and imaging can take place in the visual domain.
this caused me to think of computer graphics chipmakers nvidea, and their old slogan, "the display is the computer". and ponder the significance that visual processing is likely to make in future computer design - most especially in artificial intelligence. visual processing has been selected by nature as such a powerful tool for coping with our reality. from the series of differential equations a cheetah "intuitively" solves to intercept its prey - to even more awesome maneuvers performed by little birds twittering by us day in and day out. visual processing is hugely powerful.
i'm just saying, buy stock in nvidia.
Labels: visualization
posted by P3T3RK3Y5 for meltingearth
1 Comments:
Well, it certainly is cheap.
Post a Comment
Links to this post:
Create a Link
<< Home