but I find their solution harder to read. Too much white space, too much jumping around. Maybe I've just managed to train myself on block text by reading hundreds of novels since I was 10. Either way this is a terrible idea simply cause the information density sucks. Maybe if they threw in pictures of psychadelic kittens and badgers or something...
The real problem is that there is a generational shift in attitudes and opinions in the US. And it scares the shit out of the older crowd. People have more sex, drink more freely and are generally more open about everything. Honestly, if I were a parent I'd reserve judgment for those teaching my kids based upon my childrens progress and learning. It scares me that so many people think that children should somehow get their morals from people they'll see actively for maybe a year or two. Any kid that bases his decision to drink on what his teacher did in a photo on myspace deserves the hangover.
On a side note my 5th grade teacher told us about how he used to hunt jackrabbits from the back of a speeding pick up truck with his ex military brother. God that would've been a picture to see. A 35+ year old 5th grade teacher in the bed of a pick up with a high powered rifle. I wonder if that would've gotten his teaching license revoked. That sort of stunt could lead to far worse than drinking before you're 21.
Every time I see an 'animal rights' story it always strikes me that the people involved are doing it wrong. They always make an attempt to appeal to the 'awww don't hurt the cute widdle animal' factor and frankly most people don't care. They don't care about the 'cute bunnies' getting their heads sliced open since they're told that the animal dies so a human can live etc etc. You want to champion animal rights? Lobby for factory farm environmental oversight. Point out that these farms destroy waterways and kill the land around them. People get touchy about water supplies.
Similarly, no a chimpanzee is not a human and doing something like this trivializes the problem in the eyes of the public. You want to stop people experimenting on chimps and monkeys and other animals? Convince the public of the fact that the research doesn't hold or that it slows down research. Sometimes that is the end result of the animal research. Aspirin kills a bunch of common testing animals, the original artificial heart worked fine in dogs but the first human to get it died screaming. These sorta things make way more impact than... the conditions these animals live in suck. Most people don't care or feel that the medicine is a greater good. Certainly even in medical testing the situations should be as humane as possible, but I fail to see how this court case makes anything other than a farce of these peoples goals. Oh and stop demonizing the people who do the animal experiments. Most of them are not monsters. I say most simply cause a small percentage of humans in general are just sociopaths and they cover the gamut.
If you can't guess from the semi poorly written screed (it's late) I actually do support a fair bit of the animal rights cause but not for the save the bunny reasons. I'm just sick of people dumping antibiotics into the environment and polluting water ways and doing bad science. And stunts like this just make it harder to be taken seriously. goddamn hippies.
The term paper as such doesn't need to be done away with. What needs to be happening is that over the course of the 3 or 4 weeks the students are writing the paper and researching the paper they should also be giving a series of presentations and/or meeting with the professor at least 1 or 2 times to discuss their progress. Sure, it may sound like babying the student, but status reports are a fact of life. Hell I meet with my adviser(grad school) at least once a week just to touch base and let him know what's happening. Since this is/. as an example when I was taking computer architecture we had to do a paper on some given facet of the field or a specific architecture. Really whatever most interested us at the time. We had to provide references and brief status reports, and give a presentation on the paper at the end. You might be able to fake a paper you turn in but it's much harder to fake the presentation and the status reports if you don't actually know the material. Of course this all presumes the professor cares/has the energy to deal with this level of effort.
I'm sure many have wondered the same thing but.... what frequency band are the chips working on? I'm not a wireless guru but at the same time the profound amount of wireless interference in our world would have to be a problem for these....
On the same note I've read about short range wireless connections that are almost touching through the skin that are being considered for limb replacement so maybe it's the same tech? Of course it being late I can't remember the reference for that at the moment or how far along the trials are. At the same time this is a great approach because you don't have to worry about putting a lense system inside the ocular cavity since then you'd need to attach a number of muscular systems to the cameras which would be harder than figuring out the neural processing the eye does(from my understanding of these things) Nor do you have to have external cable ports which are just begging for trouble since they're open wounds in many ways even if they aren't bleeding.
Further side note... Dr. Boahen and his lab have done some work in modeling the behavior of the axon processing that occurs in the retina in silicon with additional processing layers in their silicon retina chips... and for the poster who wanted to know what all the extra photo receptors in the retina are for.. alot of that is for color detection and light adaptation. We actually perceive differences in signals.. so there are cells that fire more strongly in response to red but are dimmed when shown green in an antagonistic sort of effect and vice versa. Um... that's not exactly the best description in the world... Kandel Schwartz and Jessel "Principles of Neural Science" have several nice and easy to read chapters on the eyes for a nice quick intro. Probably not the most up to date work but eyes are pretty easy to understand since they're easy to get at. Brains are alot harder. To my understanding of what I've read(I'm not a neural scientist but I do study computer vision and biological inspiration is fun) most of the spatial and motion processing occurs in the Visual Cortex pathways more so than the eye. Please correct me if I'm mistaken but that's my understanding of things.
1) They're not researching neural networks in the classic sense of AI research. They're not trying to come up with an approximate model. Rather they're trying to just design a neuronal test bed and connect it similarly to the brain and see what happens. No software models to speak of really.
2) Writing software to model these accurately is actually much harder than just doing it in hardware due to the massively parallel nature of the computations and the neural connections. They aren't just creating layers and doing back propagation like you would in a more standard NN.
3) Sometimes it's just better to do somehing the hard and right way than it is to try to build things up in stages. Further, it's not like they probably won't design the chips(or have designed the chips already) in software layout tools and simulate the hell out of them there.
Having been a fan of neuromorphic engineering for several years now(Note I'm not an active researcher but I pretend somedays:) ) one of the major advantages of neuromorphic functionality isn't necessarily it's ability to model biological systems but the fact that the devices are extremely low power. When modeling neurons in silicon(at least back in the day of Carver Mead's work and for cochlea and retina stuff and I'm doubting it's changed too bunch but I could be wrong) the transistors would run in sub threshhold mode(basically leakage currents so OFF) since the power curves modeled the expected neuro response curves. One of Boahen's stated goals(at least on his website when he was at Penn) was to reduce power consumption and improve processing power for problem solving via these techniques. His lab has been in Scientific America a couple times in the last few years for work in accurately modeling Neuronal spiking in hardware too. I have them but not at hand so I can't cite them at the moment but they were fun reads.
So in summary, it's more than just modeling the brain. It's about letting biology inspire us to make better and more efficient computing systems.
Welp while a certain level of anger and rage is present in said post. I've gotta say... I agree. Religions are the problem. And I'm sick and tired of this attitude of "Oh we have to respect your religion even if you're attempting to forcefeed it to us or kill us for being infidels." When the Muslims rioted and burned cars and buildings and Embassies, Europe and the US both sat on their hands. We tried to placate these people(religous whackjobs.. not muslims in particular). At the point in time when people are BURNING embassies it should be pretty damn obvious that we're in a cultural war between religion and... well... the rest of us. I've gotta say I believe in many of the "Western" world ideals. I could do with a bit less greed and a bit more compassion for ones fellow man but the basic premise of things is on the right track. As a simple example... in most of europe and the americas women get a much much fairer shake than they do in the more religious communities(well mainly religions of the "book" aka judaism, christianity and islam). THIS IS A GOOD THING(tm). Sure the western world is flawed but we also don't have female circumcisions, women forced to wear covers with penalty of death or mobs of people killing others for things they don't like with impugnity. I'm not religious... even if I were a christian I still wouldn't be religious. Belief\faith in something is fine... whatever gets you through the day... but when you start telling me what has to get me through the day it's time to fight. Too many people are just willing to let these religious zealots and bullies roll over them or to sit quietly and hope the problem will just go away. It won't and we need to stop placating people. Tell them to shut up.. Tell them to get the hell out of schools, government, politics or whatever the hell else they want to control.
Just for the sake of fairness when the orthodox jews recently rioted about a gay pride parade they should have been rounded up and charged to the fullest extent of the law for destruction of property and disturbing the peace. I mean hell last year they stabbed someone for being homosexual.. How crazy is that.. I mean really what right does anyone have to determine your sexuality... If you're not sexually involved with that person... the answer is none.
I know this is a bit late in the game to even get seen on slashdot but if you start tracing the evangical christian right's role in the bush admin and lately even republican parties you'll start to see a driving goal towards theocracy. Bush has always had support from and worked closely with the evangalists in power. Focus on the Family leaders brag that they can control votes simply with a phone call since they have nearly 40 some million voters behind them and about 70million people total. That's right, close to a third of the country, buys into the ideas that the government should incorporate God heavily. Maybe I'm all fired up cause I just got done reading a bunch of stuff about this movement(and actually believe in the seperation of church and state) but the roots of this campaign towards power start in the 70's actually with groups like the heritage thinktank and others. Mostly founded by weinrych(think that's the spelling) with money from Coors. They down play the role heavily but it all got started as an attempt to combat the grass roots momentum of the Democratic party in the 60's. This isn't an attack on religion but regilion is staging an attack on the rest of society. They have millions of dollars which are tax free since the are churches( and nominally non political) but they spend a large amount of time "getting the vote out". In fact Focus on the Family has set up a front for their political actions so they can mask the money in the church side and formally shill for the party they want.
Ask yourself if you really think it's a good idea that currently 39 senators in power have voted 100% the way the Christian Coalition would've wanted them to on the issues that concern it. Not to seem too out there but everytime I read about them (or seen them trying to intervene in local matters in my hometown) i've always just sat there thinking to myself "wow, just wow these people are crazy"
Okies I'm done ranting but if you're concerned about the state of this country and don't want it to become a giant morass of a theocracy you really should go read about these people and their agendas. They've been working on this for 30 years and they're crazy.
I mentioned the rationals more to illustrate the concept that being defined on all data sets and rigour have nothing to do with one another. Nothing more to it than that. I mentioned that you can treat points of local maximum in your gradient function as edge points. Beyond that the simplest definition is that an edge point is a point of discontinuity(hence the idea of maximizing the gradient locally). The example you cited is a valid signal to consider. However every point would be a point of discontinuity and thus be an edge point. To be complete we can define edges as sets consisting of edge points which are connected or close under a specified distance metric. Beyond simple differentiation, specific edge detectors choose to do different things to help extract edges which are optimal or dominant according to the desired operating parameters specified by their construction. In the context of the Canny edge detector your proposed signal would cause difficulties not because of the underlying theory but due to other operating parameters that the detector is working under i.e. the detector tries to enusure that detected edge points are localized to edges that support them.
As to defining noise, noise is nebulous. Noise is just undesired signal activity or signal corruption. Therefore a definition of noise is specific to every application. I said your example was essentially noise to an edge detector since the detector would detect every point which would defeat the purpose of using a detector since the hope of an edge detector is to reduce the information one must consider. In this sense I was defining noise to be a signal for which the edge detector gave no meaningful data in the context of edge detection i.e. everything is an edge point but nothing is an edge(under the assumption that you're operating with a metric which says edge points are close if they are spatially related and if the gradient function is close in value). As for more "acceptable" images, one needs only consider images or signals which vary at a slower rate or have only a sparser number of discontinuities. Your example wasn't unworkable just more of a worse case scenerio where the behavior of an edge detector wouldn't be useful in practical terms.
I have an undergraduate degree in Computer Science and Mathematics, I've also got a minors in image processing and electrical engineering, I've almost completed my CS masters, I'm working on an applied math masters and a CS phd. Just to establish my credentials on this one. I get what rigour is . Mathematical rigour does not imply that a function is defined in all contexts. The example you propose is a degenerate one since I'm assuming you want a matrix of alternating 1's and 0's and not an actual block matrix of black and white squares such as you'd use to calibrate a camera. That matrix is the equivalent of noise from a derivative standpoint. It falls into the same category as the function of marking the rationals with 1's and irrationals with 0's and asking what's the integral of this function. Reimann integration can't tell you this(hence by your definition isn't rigourous). Measure theory will tell you it's 0. That's beyond the scope of the edge detector thing but I like the example.
As far as the result of the noise matrix you propose, the gradient in the X or Y direction of the image is defined as a finite element difference. A well known numerical technique. You could also use a number of other filters to extract gradient information. Whether that information is useful is beyond the scope of the question but it is defined to operate there. A Canny detector would freak out to a degree on that image since it would attempt to smooth the noise into continuous edges which was one of the goals Canny built directly into his detector. If you look at his work he set out to accomplish a set of specific goals when given an image(under reasonable constraints) and then examined the optimization problem these goals created. He then determined what an optimal detector would look like but then approximated it to make computation more efficient. If you read a book on image processing such as Gonzales and Woods introductory text you will see that image processing is indeed founded on the principles of rigourous analysis. The problems you are complaining about are the result of the fact that digital images are discrete. Discrete sample spaces play havoc with all the nice continous math behind the scenes and I suppose a number of what you would call hacks become necessary to develop any theory.
As for your final point of wanting a definition of an edge, an edge is a boundary between two objects. Those objects can be surfaces, color patches, texture regions. Pretty much anything you'd like. If I draw a curve in the plane, say a circle, and I ask you where the boundary between the inside and the outside of the circle lies(a defined question provided you give the curve an orientation and apply the Jordan curve theorem(Been a while but i'm pretty sure Jordan is the guy)) you'd be willing to say the circle divides these regions, ie circle points are edge points. In particular, Canny defined a point as being an edge if the graiant norm reached a local maximum at that point. I hate to break it to you but that's what mathimaticl rigour is. You can define something as having the functionality you want and then determine the results of that in a rigourous manner. Admittedly you wish to work in some sense from a minimal definition so as to not make too many assumptions in one place.
I do agree with you though that many CV practioners do not subscribe to enough mathamatical rigour and that many times poor ideas are introduced but eigenfaces are not mathematically ill-defined. I'm not that fond of them but they do form a basis for the face space as determined by a PCA framework.
I'm afraid I'm going to call shennanigans on some of this. I've been doing Vision work for about 5 years now with a hefty does of image and signal processing in the mix(Working as gradstudent in the field right now in fact). Edge detection is well defined. The canny and shah-istan(think that's the name) are about as close to a mathematical optimal edge detector as one can get. There is in fact a well developed body of theory regarding differentiation of Signals. The problem doesn't lie in the mathematical models involved. It lies in how many people want to use those models. Edge detection suffers from spurious edges or edge flakes which are a symptom of noise in the signal at differention(ie differentiation enhances noise, integration smooths it).
Segmentation can also be well defined you just have to be clear on what it is you're segmenting. Are you working in a color space, texture, motion? That matters. However you can get some very good results in these fields. See GPCA techniques for some examples of doing it. Or even modified PCA + EM or PCA+ Kmeans(clustering theory). Again very well defined.
Mathematically there are several models for face recognition. One can examine the ideas of eigen faces(not my personal favorite but it's there), kernel based SSD type approaches to find key points, partial face detection followed by recognition over a sequence of images used to reconstruct the face, and more.
The problem isn't the math. It's that when you project a model you are essentially destroying an entire degree of freedom which is a huge deal. Further just as you can match a partial finger print or a partial ear print you can match partial facechunks. The problem with makeup or facial hair comes when one relies on global matching techniques or uses only 2d information to do the matching.
Now I'll be a first to say that alot of computer vision is a solution in search of a problem or that people do use a number of cheap hacks and dirty tricks to get things working but saying it's not mathematical is a lie. I can turn around and see at least 3 books at a glance that detail the mathematics that are a part of vision and image processing. So please don't confuse peoples fuzzy use or lack of understanding of the math for there being no math.
Note: Machines are also bad at a number of tasks humans are really good at but the same can be said that there are many tasks that humans are very bad at but the machines excel at. Absolute range detection is a good example. Humans are very bad at telling you the exact range to an object, even with some sort of scale of the scene reference. Computers on the other hand(while suffering from noise in the signal) are still able to achieve significant accuracy depending on the range. You can see tyzx for an example of a comany who makes highly accurate stereo rigs.(They were around as of 2 years ago at least and I assume they're still going strong)
Cheers
Yah, But i'm gonna go out on a limb and say that Mulholland Drive could've only been made better had a commitee been involved. As was it was childish and silly and just a bit stupid over all. Then again I did watch it twice... course the first viewing was in a really creepy theatre with chairs that prevented actually thinking about the movie due to the pain of sitting in them. Terrible film.
um.... the nuclear targets in Japan during WW2 also happened to be manufacturing centers for the Japanese army.
Also if you want to get down to it the Nukes weren't even that destructive compared to oh say the firebombing raids of tokyo which destroyed about 16 square miles of tokyo in a firestorm so intense it broke some of the bombers into pieces from the winds it created.
Pure reactionary nonesense... just cause the nukes were bigger and flashier does not mean they were the most destructive. And no no one said "LETS GO KILL A BUNCH OF NIP CIVILIANS THAT'LL SHOW THEM"... well maybe some people did but that wasn't the logic used in picking the targets.
Europe annouces a space billboard initiative. Part of this initiative involves a unilateral declaration that any attempt to remove their billboards will be seen as an act of agression. Followed by what sounded like muffled laughter.
I just finished a course on this particular subject actually. A few fun comments.
In one biological study in Maine (Great Duck Island) it seems that the birds on the island they were monitoring had been attacking the sensor motes. In another case these devices offered the first look ever at night time migration patterns of zebras.(aka ZebraNet)
As far as military applications go the one that I am most aware of is DARPA's sniper net. It's a system of audio sensors designed to locate and pinpoint snipers based on gunshot triangulation.
There are some earthquake structural monitoring systems being built in California as well.
However things to be cautioned about. The smaller motes do not have very much in the way of processing power(ie can't even do floating point) so there's no need to get really paranoid about secret cameras. Most of the motes with cameras are big enough you'd probably notice them if you were looking. Primarily the motes are equipped with various sensor banks for things such as Light, Temp, Vibration, Audio, etc etc. Also if you're interested in working with the software for these things the primary OS people use is TinyOS. However a word of caution, if you want to muck around inside the inner workings of TinyOS you're pretty much on your own and some of the things are already legacy. The coolest part of sensor nets, in my opinion,is the ability to do in network data processing as the data is funneled through the network. Oh and there's already a Database system designed for use in these systems. It's name is TinyDB(surprising naming scheme I know)
I actually saw a talk by one of the people involved in the work being done at Berkeley on this very subject Thursday. His names Robert J. Full and he's with the Biology dept at Berkeley. His and his students work have inspired an entire host of materials engineering and robotics applications. For instance his work with geckos has lead to work being done on producing nanoscale nonsticky adhesives which are based on van der Waal forces. Also based on his studies of the cockroach(and again his students) we're seeing some robots which are capable of climbing up walls and transversing really really rough terrain. Definitely someone to keep track of if you're interested in biological inspiration for engineering.
Actually, researchers at Johns Hopkins developed flexible wires that can flex and stretch upto 1.5 times their own length without suffering damage. The article was in Nature back in March, but you need to have a subscription to view it now.
Slashdot Mirror
but I find their solution harder to read. Too much white space, too much jumping around. Maybe I've just managed to train myself on block text by reading hundreds of novels since I was 10. Either way this is a terrible idea simply cause the information density sucks. Maybe if they threw in pictures of psychadelic kittens and badgers or something...
The real problem is that there is a generational shift in attitudes and opinions in the US. And it scares the shit out of the older crowd. People have more sex, drink more freely and are generally more open about everything. Honestly, if I were a parent I'd reserve judgment for those teaching my kids based upon my childrens progress and learning. It scares me that so many people think that children should somehow get their morals from people they'll see actively for maybe a year or two. Any kid that bases his decision to drink on what his teacher did in a photo on myspace deserves the hangover.
On a side note my 5th grade teacher told us about how he used to hunt jackrabbits from the back of a speeding pick up truck with his ex military brother. God that would've been a picture to see. A 35+ year old 5th grade teacher in the bed of a pick up with a high powered rifle. I wonder if that would've gotten his teaching license revoked. That sort of stunt could lead to far worse than drinking before you're 21.
Every time I see an 'animal rights' story it always strikes me that the people involved are doing it wrong. They always make an attempt to appeal to the 'awww don't hurt the cute widdle animal' factor and frankly most people don't care. They don't care about the 'cute bunnies' getting their heads sliced open since they're told that the animal dies so a human can live etc etc. You want to champion animal rights? Lobby for factory farm environmental oversight. Point out that these farms destroy waterways and kill the land around them. People get touchy about water supplies.
Similarly, no a chimpanzee is not a human and doing something like this trivializes the problem in the eyes of the public. You want to stop people experimenting on chimps and monkeys and other animals? Convince the public of the fact that the research doesn't hold or that it slows down research. Sometimes that is the end result of the animal research. Aspirin kills a bunch of common testing animals, the original artificial heart worked fine in dogs but the first human to get it died screaming. These sorta things make way more impact than... the conditions these animals live in suck. Most people don't care or feel that the medicine is a greater good. Certainly even in medical testing the situations should be as humane as possible, but I fail to see how this court case makes anything other than a farce of these peoples goals. Oh and stop demonizing the people who do the animal experiments. Most of them are not monsters. I say most simply cause a small percentage of humans in general are just sociopaths and they cover the gamut.
If you can't guess from the semi poorly written screed (it's late) I actually do support a fair bit of the animal rights cause but not for the save the bunny reasons. I'm just sick of people dumping antibiotics into the environment and polluting water ways and doing bad science. And stunts like this just make it harder to be taken seriously. goddamn hippies.
The term paper as such doesn't need to be done away with. What needs to be happening is that over the course of the 3 or 4 weeks the students are writing the paper and researching the paper they should also be giving a series of presentations and/or meeting with the professor at least 1 or 2 times to discuss their progress. Sure, it may sound like babying the student, but status reports are a fact of life. Hell I meet with my adviser(grad school) at least once a week just to touch base and let him know what's happening. Since this is /. as an example when I was taking computer architecture we had to do a paper on some given facet of the field or a specific architecture. Really whatever most interested us at the time. We had to provide references and brief status reports, and give a presentation on the paper at the end. You might be able to fake a paper you turn in but it's much harder to fake the presentation and the status reports if you don't actually know the material. Of course this all presumes the professor cares/has the energy to deal with this level of effort.
I'm sure many have wondered the same thing but.... what frequency band are the chips working on? I'm not a wireless guru but at the same time the profound amount of wireless interference in our world would have to be a problem for these....
On the same note I've read about short range wireless connections that are almost touching through the skin that are being considered for limb replacement so maybe it's the same tech? Of course it being late I can't remember the reference for that at the moment or how far along the trials are. At the same time this is a great approach because you don't have to worry about putting a lense system inside the ocular cavity since then you'd need to attach a number of muscular systems to the cameras which would be harder than figuring out the neural processing the eye does(from my understanding of these things) Nor do you have to have external cable ports which are just begging for trouble since they're open wounds in many ways even if they aren't bleeding.
Further side note... Dr. Boahen and his lab have done some work in modeling the behavior of the axon processing that occurs in the retina in silicon with additional processing layers in their silicon retina chips... and for the poster who wanted to know what all the extra photo receptors in the retina are for.. alot of that is for color detection and light adaptation. We actually perceive differences in signals.. so there are cells that fire more strongly in response to red but are dimmed when shown green in an antagonistic sort of effect and vice versa. Um... that's not exactly the best description in the world... Kandel Schwartz and Jessel "Principles of Neural Science" have several nice and easy to read chapters on the eyes for a nice quick intro. Probably not the most up to date work but eyes are pretty easy to understand since they're easy to get at. Brains are alot harder. To my understanding of what I've read(I'm not a neural scientist but I do study computer vision and biological inspiration is fun) most of the spatial and motion processing occurs in the Visual Cortex pathways more so than the eye. Please correct me if I'm mistaken but that's my understanding of things.
1) They're not researching neural networks in the classic sense of AI research. They're not trying to come up with an approximate model. Rather they're trying to just design a neuronal test bed and connect it similarly to the brain and see what happens. No software models to speak of really.
2) Writing software to model these accurately is actually much harder than just doing it in hardware due to the massively parallel nature of the computations and the neural connections. They aren't just creating layers and doing back propagation like you would in a more standard NN.
3) Sometimes it's just better to do somehing the hard and right way than it is to try to build things up in stages. Further, it's not like they probably won't design the chips(or have designed the chips already) in software layout tools and simulate the hell out of them there.
Having been a fan of neuromorphic engineering for several years now(Note I'm not an active researcher but I pretend somedays :) ) one of the major advantages of neuromorphic functionality isn't necessarily it's ability to model biological systems but the fact that the devices are extremely low power. When modeling neurons in silicon(at least back in the day of Carver Mead's work and for cochlea and retina stuff and I'm doubting it's changed too bunch but I could be wrong) the transistors would run in sub threshhold mode(basically leakage currents so OFF) since the power curves modeled the expected neuro response curves. One of Boahen's stated goals(at least on his website when he was at Penn) was to reduce power consumption and improve processing power for problem solving via these techniques. His lab has been in Scientific America a couple times in the last few years for work in accurately modeling Neuronal spiking in hardware too. I have them but not at hand so I can't cite them at the moment but they were fun reads.
So in summary, it's more than just modeling the brain. It's about letting biology inspire us to make better and more efficient computing systems.
Welp while a certain level of anger and rage is present in said post. I've gotta say... I agree. Religions are the problem. And I'm sick and tired of this attitude of "Oh we have to respect your religion even if you're attempting to forcefeed it to us or kill us for being infidels." When the Muslims rioted and burned cars and buildings and Embassies, Europe and the US both sat on their hands. We tried to placate these people(religous whackjobs.. not muslims in particular). At the point in time when people are BURNING embassies it should be pretty damn obvious that we're in a cultural war between religion and... well... the rest of us. I've gotta say I believe in many of the "Western" world ideals. I could do with a bit less greed and a bit more compassion for ones fellow man but the basic premise of things is on the right track. As a simple example... in most of europe and the americas women get a much much fairer shake than they do in the more religious communities(well mainly religions of the "book" aka judaism, christianity and islam). THIS IS A GOOD THING(tm). Sure the western world is flawed but we also don't have female circumcisions, women forced to wear covers with penalty of death or mobs of people killing others for things they don't like with impugnity. I'm not religious... even if I were a christian I still wouldn't be religious. Belief\faith in something is fine... whatever gets you through the day... but when you start telling me what has to get me through the day it's time to fight. Too many people are just willing to let these religious zealots and bullies roll over them or to sit quietly and hope the problem will just go away. It won't and we need to stop placating people. Tell them to shut up.. Tell them to get the hell out of schools, government, politics or whatever the hell else they want to control.
Just for the sake of fairness when the orthodox jews recently rioted about a gay pride parade they should have been rounded up and charged to the fullest extent of the law for destruction of property and disturbing the peace. I mean hell last year they stabbed someone for being homosexual.. How crazy is that.. I mean really what right does anyone have to determine your sexuality... If you're not sexually involved with that person... the answer is none.
I know this is a bit late in the game to even get seen on slashdot but if you start tracing the evangical christian right's role in the bush admin and lately even republican parties you'll start to see a driving goal towards theocracy. Bush has always had support from and worked closely with the evangalists in power. Focus on the Family leaders brag that they can control votes simply with a phone call since they have nearly 40 some million voters behind them and about 70million people total. That's right, close to a third of the country, buys into the ideas that the government should incorporate God heavily. Maybe I'm all fired up cause I just got done reading a bunch of stuff about this movement(and actually believe in the seperation of church and state) but the roots of this campaign towards power start in the 70's actually with groups like the heritage thinktank and others. Mostly founded by weinrych(think that's the spelling) with money from Coors. They down play the role heavily but it all got started as an attempt to combat the grass roots momentum of the Democratic party in the 60's. This isn't an attack on religion but regilion is staging an attack on the rest of society. They have millions of dollars which are tax free since the are churches( and nominally non political) but they spend a large amount of time "getting the vote out". In fact Focus on the Family has set up a front for their political actions so they can mask the money in the church side and formally shill for the party they want.
Ask yourself if you really think it's a good idea that currently 39 senators in power have voted 100% the way the Christian Coalition would've wanted them to on the issues that concern it. Not to seem too out there but everytime I read about them (or seen them trying to intervene in local matters in my hometown) i've always just sat there thinking to myself "wow, just wow these people are crazy"
Okies I'm done ranting but if you're concerned about the state of this country and don't want it to become a giant morass of a theocracy you really should go read about these people and their agendas. They've been working on this for 30 years and they're crazy.
I mentioned the rationals more to illustrate the concept that being defined on all data sets and rigour have nothing to do with one another. Nothing more to it than that. I mentioned that you can treat points of local maximum in your gradient function as edge points. Beyond that the simplest definition is that an edge point is a point of discontinuity(hence the idea of maximizing the gradient locally). The example you cited is a valid signal to consider. However every point would be a point of discontinuity and thus be an edge point. To be complete we can define edges as sets consisting of edge points which are connected or close under a specified distance metric. Beyond simple differentiation, specific edge detectors choose to do different things to help extract edges which are optimal or dominant according to the desired operating parameters specified by their construction. In the context of the Canny edge detector your proposed signal would cause difficulties not because of the underlying theory but due to other operating parameters that the detector is working under i.e. the detector tries to enusure that detected edge points are localized to edges that support them.
As to defining noise, noise is nebulous. Noise is just undesired signal activity or signal corruption. Therefore a definition of noise is specific to every application. I said your example was essentially noise to an edge detector since the detector would detect every point which would defeat the purpose of using a detector since the hope of an edge detector is to reduce the information one must consider. In this sense I was defining noise to be a signal for which the edge detector gave no meaningful data in the context of edge detection i.e. everything is an edge point but nothing is an edge(under the assumption that you're operating with a metric which says edge points are close if they are spatially related and if the gradient function is close in value). As for more "acceptable" images, one needs only consider images or signals which vary at a slower rate or have only a sparser number of discontinuities. Your example wasn't unworkable just more of a worse case scenerio where the behavior of an edge detector wouldn't be useful in practical terms.
pissing contest mode.
/pissing contest mode
I have an undergraduate degree in Computer Science and Mathematics, I've also got a minors in image processing and electrical engineering, I've almost completed my CS masters, I'm working on an applied math masters and a CS phd. Just to establish my credentials on this one. I get what rigour is . Mathematical rigour does not imply that a function is defined in all contexts. The example you propose is a degenerate one since I'm assuming you want a matrix of alternating 1's and 0's and not an actual block matrix of black and white squares such as you'd use to calibrate a camera. That matrix is the equivalent of noise from a derivative standpoint. It falls into the same category as the function of marking the rationals with 1's and irrationals with 0's and asking what's the integral of this function. Reimann integration can't tell you this(hence by your definition isn't rigourous). Measure theory will tell you it's 0. That's beyond the scope of the edge detector thing but I like the example.
As far as the result of the noise matrix you propose, the gradient in the X or Y direction of the image is defined as a finite element difference. A well known numerical technique. You could also use a number of other filters to extract gradient information. Whether that information is useful is beyond the scope of the question but it is defined to operate there. A Canny detector would freak out to a degree on that image since it would attempt to smooth the noise into continuous edges which was one of the goals Canny built directly into his detector. If you look at his work he set out to accomplish a set of specific goals when given an image(under reasonable constraints) and then examined the optimization problem these goals created. He then determined what an optimal detector would look like but then approximated it to make computation more efficient. If you read a book on image processing such as Gonzales and Woods introductory text you will see that image processing is indeed founded on the principles of rigourous analysis. The problems you are complaining about are the result of the fact that digital images are discrete. Discrete sample spaces play havoc with all the nice continous math behind the scenes and I suppose a number of what you would call hacks become necessary to develop any theory.
As for your final point of wanting a definition of an edge, an edge is a boundary between two objects. Those objects can be surfaces, color patches, texture regions. Pretty much anything you'd like. If I draw a curve in the plane, say a circle, and I ask you where the boundary between the inside and the outside of the circle lies(a defined question provided you give the curve an orientation and apply the Jordan curve theorem(Been a while but i'm pretty sure Jordan is the guy)) you'd be willing to say the circle divides these regions, ie circle points are edge points. In particular, Canny defined a point as being an edge if the graiant norm reached a local maximum at that point. I hate to break it to you but that's what mathimaticl rigour is. You can define something as having the functionality you want and then determine the results of that in a rigourous manner. Admittedly you wish to work in some sense from a minimal definition so as to not make too many assumptions in one place.
I do agree with you though that many CV practioners do not subscribe to enough mathamatical rigour and that many times poor ideas are introduced but eigenfaces are not mathematically ill-defined. I'm not that fond of them but they do form a basis for the face space as determined by a PCA framework.
I'm afraid I'm going to call shennanigans on some of this. I've been doing Vision work for about 5 years now with a hefty does of image and signal processing in the mix(Working as gradstudent in the field right now in fact). Edge detection is well defined. The canny and shah-istan(think that's the name) are about as close to a mathematical optimal edge detector as one can get. There is in fact a well developed body of theory regarding differentiation of Signals. The problem doesn't lie in the mathematical models involved. It lies in how many people want to use those models. Edge detection suffers from spurious edges or edge flakes which are a symptom of noise in the signal at differention(ie differentiation enhances noise, integration smooths it). Segmentation can also be well defined you just have to be clear on what it is you're segmenting. Are you working in a color space, texture, motion? That matters. However you can get some very good results in these fields. See GPCA techniques for some examples of doing it. Or even modified PCA + EM or PCA+ Kmeans(clustering theory). Again very well defined. Mathematically there are several models for face recognition. One can examine the ideas of eigen faces(not my personal favorite but it's there), kernel based SSD type approaches to find key points, partial face detection followed by recognition over a sequence of images used to reconstruct the face, and more. The problem isn't the math. It's that when you project a model you are essentially destroying an entire degree of freedom which is a huge deal. Further just as you can match a partial finger print or a partial ear print you can match partial facechunks. The problem with makeup or facial hair comes when one relies on global matching techniques or uses only 2d information to do the matching. Now I'll be a first to say that alot of computer vision is a solution in search of a problem or that people do use a number of cheap hacks and dirty tricks to get things working but saying it's not mathematical is a lie. I can turn around and see at least 3 books at a glance that detail the mathematics that are a part of vision and image processing. So please don't confuse peoples fuzzy use or lack of understanding of the math for there being no math. Note: Machines are also bad at a number of tasks humans are really good at but the same can be said that there are many tasks that humans are very bad at but the machines excel at. Absolute range detection is a good example. Humans are very bad at telling you the exact range to an object, even with some sort of scale of the scene reference. Computers on the other hand(while suffering from noise in the signal) are still able to achieve significant accuracy depending on the range. You can see tyzx for an example of a comany who makes highly accurate stereo rigs.(They were around as of 2 years ago at least and I assume they're still going strong) Cheers
Yah, But i'm gonna go out on a limb and say that Mulholland Drive could've only been made better had a commitee been involved. As was it was childish and silly and just a bit stupid over all. Then again I did watch it twice... course the first viewing was in a really creepy theatre with chairs that prevented actually thinking about the movie due to the pain of sitting in them. Terrible film.
um.... the nuclear targets in Japan during WW2 also happened to be manufacturing centers for the Japanese army.
Also if you want to get down to it the Nukes weren't even that destructive compared to oh say the firebombing raids of tokyo which destroyed about 16 square miles of tokyo in a firestorm so intense it broke some of the bombers into pieces from the winds it created.
Pure reactionary nonesense... just cause the nukes were bigger and flashier does not mean they were the most destructive. And no no one said "LETS GO KILL A BUNCH OF NIP CIVILIANS THAT'LL SHOW THEM"... well maybe some people did but that wasn't the logic used in picking the targets.
Europe annouces a space billboard initiative. Part of this initiative involves a unilateral declaration that any attempt to remove their billboards will be seen as an act of agression. Followed by what sounded like muffled laughter.
I just finished a course on this particular subject actually. A few fun comments.
In one biological study in Maine (Great Duck Island) it seems that the birds on the island they were monitoring had been attacking the sensor motes. In another case these devices offered the first look ever at night time migration patterns of zebras.(aka ZebraNet)
As far as military applications go the one that I am most aware of is DARPA's sniper net. It's a system of audio sensors designed to locate and pinpoint snipers based on gunshot triangulation.
There are some earthquake structural monitoring systems being built in California as well.
However things to be cautioned about. The smaller motes do not have very much in the way of processing power(ie can't even do floating point) so there's no need to get really paranoid about secret cameras. Most of the motes with cameras are big enough you'd probably notice them if you were looking. Primarily the motes are equipped with various sensor banks for things such as Light, Temp, Vibration, Audio, etc etc. Also if you're interested in working with the software for these things the primary OS people use is TinyOS. However a word of caution, if you want to muck around inside the inner workings of TinyOS you're pretty much on your own and some of the things are already legacy. The coolest part of sensor nets, in my opinion,is the ability to do in network data processing as the data is funneled through the network. Oh and there's already a Database system designed for use in these systems. It's name is TinyDB(surprising naming scheme I know)
Cheers
I actually saw a talk by one of the people involved in the work being done at Berkeley on this very subject Thursday. His names Robert J. Full and he's with the Biology dept at Berkeley. His and his students work have inspired an entire host of materials engineering and robotics applications. For instance his work with geckos has lead to work being done on producing nanoscale nonsticky adhesives which are based on van der Waal forces. Also based on his studies of the cockroach(and again his students) we're seeing some robots which are capable of climbing up walls and transversing really really rough terrain. Definitely someone to keep track of if you're interested in biological inspiration for engineering.
Actually, researchers at Johns Hopkins developed flexible wires that can flex and stretch upto 1.5 times their own length without suffering damage. The article was in Nature back in March, but you need to have a subscription to view it now.