Unfortunately no-- since I developed them at work and signed (foolishly) my soul over to the company, they own all of it and I would have to go through a tedious process to extract it legally...
I've recently done quite a bit of research on Neural Networks, including coding and simulating them by hand... There are some (qutie drastic) flaws with neural networks...
I started my research doing a classic 5 pixel by 5 pixel OCR (optical character recognition) on the domain of digits on a single layer perceptron type network (similar to what these guys were using minuns the delayed firing rate)
Not suprisingly, the training algorithm converged to an answer quite quickly and I proceded to run tests with noisy data, to test the genrealazation of the network.
100 per cent correct at zero noise
50 per cent correct at twnety-five per cent nosie
10 per cent correct at fifty per cent noise
NEARLY zero percent correct above fifty.
This isn't shocking in itself until you realize that once you go above fifty percent distortion rates you are actually INVERTING the digit!
I retrained the network with inverted digits as well as the normal digits and re-ran the tests on the same set of data (note: The net WILL NOT converge on normal & inverted 5x5 digits with only ten cells).. The correctness rate was only twnety-per cent throughout the whole domain of noise levels.
I then retrained again using TWENTY cells (9 more than this articles) and it converge quite nicely and gave me a quadratic function with an R-Squared value of.9995 or so.
People view Neural networks sometimes as a fix-all solution.. The article on/. earlier about "eveloutionary computing" is the same premise as neural networks : try stuff randomly (or using calculus) until we get a decent solution.
I'm sorry kiddoes, but that just doesn't cut it. A neural network can't ever outperform a Turing machine so there can't be any chance in hell it will ever outperform us in non-specilized tasks.
Of course, I'd probably be more optimistic if these guys would have released there algorithms, papers, source-code, etc so we could actually figure out HOW the HELL they can get an 11 cell network to recognize speech...
The moral of the story? understanding speech is a hell of lot harder than recognizing ten digits!
I happen to work as an intern at a fortune 50 company right now-- and i've been asking around... They have every spare engineer and technician working on algorithms/programs/controls etc for the automation of data analysis.
NASA and the Air Force started it years ago with a project they did to detect (and thereby save money on unschedulred repiars) early failures on rotor shafts and bearings.
I find this stuff very boring and rudimentry. The *only* reason any company keeps this propeirty is because they don't want people to know how crappy their products are (i'm referring to in-house OLAP development, where release would mean exposure to their most sensitive failure data).
IBM apparently uses OLAP on their hard drives because they don't quote their mean-time-between failures. I even called them and asked.
I've used their SAS programming language earlier in the year, and I'll tell you-- it's NO walk in the park. The syntax is worse than umm.. well.. I guess it's just the worse.
The only reasonable solution to this lad's problem would be to develop his own system.. It's the cheapest, probably the most reliable and, would be, by far, the most customizable.
I feel for him for having to do this kind of work though because I was driven mad by it. I still have a bad taste in my mouth from it.
My suggestion to him : hire a bunch of interns and have *them* do it.
I'm a home-grown American, and I can say that we don't *always* have that attitude. Our current measurement system isn't called English for our health.... >p>But the main reason we have that belief at all is because it seems everything foreign wants to blow up our government buildings ?? Maybe I'm just being paranoid...
And another thing-- If we converted to metric, I would cry for days because I wouldn't be able to coast on the expressways at 75 miles per hour.. It would be like 140 km/hr..
Your first theory cannot be correct. "it must be concentrated more densely near the sun and less densely farther away" violates one of the leading views on space and time-- that the universe looks and acts the same no matter where you are. The second theory-- If dark matter did exist, I seriously doubt it would have any effect on the space probe because A) It is too sparse B) If it weren't, we'd have found more by now And back in the day, people used the "Ether" theory to explain why light always travels the same speed and similar effects. It is my opinion that the apparent error in this probe's speed and acceleration are due to it's moving through space/time at 27,000 mph. This probably has a small effect locally, but over time it builds up and in effect, moves through time slower than you or I. Of course, that is all Einstein's general theory, and the NASA guys have already looked at that.. I would be really interested in seeing them post all of their equations and work so we can check for errors.
Slashdot Mirror
Very similar to 1984 by George Orwell.. Frightening at best
Unfortunately no-- since I developed them at work and signed (foolishly) my soul over to the company, they own all of it and I would have to go through a tedious process to extract it legally...
Sorry,
I've recently done quite a bit of research on Neural Networks, including coding and simulating them by hand... There are some (qutie drastic) flaws with neural networks...
I started my research doing a classic 5 pixel by 5 pixel OCR (optical character recognition) on the domain of digits on a single layer perceptron type network (similar to what these guys were using minuns the delayed firing rate)
Not suprisingly, the training algorithm converged to an answer quite quickly and I proceded to run tests with noisy data, to test the genrealazation of the network.
This isn't shocking in itself until you realize that once you go above fifty percent distortion rates you are actually INVERTING the digit!
I retrained the network with inverted digits as well as the normal digits and re-ran the tests on the same set of data (note: The net WILL NOT converge on normal & inverted 5x5 digits with only ten cells).. The correctness rate was only twnety-per cent throughout the whole domain of noise levels.
I then retrained again using TWENTY cells (9 more than this articles) and it converge quite nicely and gave me a quadratic function with an R-Squared value of .9995 or so.
People view Neural networks sometimes as a fix-all solution.. The article on /. earlier about "eveloutionary computing" is the same premise as neural networks : try stuff randomly (or using calculus) until we get a decent solution.
I'm sorry kiddoes, but that just doesn't cut it. A neural network can't ever outperform a Turing machine so there can't be any chance in hell it will ever outperform us in non-specilized tasks.
Of course, I'd probably be more optimistic if these guys would have released there algorithms, papers, source-code, etc so we could actually figure out HOW the HELL they can get an 11 cell network to recognize speech...
The moral of the story? understanding speech is a hell of lot harder than recognizing ten digits!
I happen to work as an intern at a fortune 50 company right now-- and i've been asking around... They have every spare engineer and technician working on algorithms/programs/controls etc for the automation of data analysis.
NASA and the Air Force started it years ago with a project they did to detect (and thereby save money on unschedulred repiars) early failures on rotor shafts and bearings.
I find this stuff very boring and rudimentry. The *only* reason any company keeps this propeirty is because they don't want people to know how crappy their products are (i'm referring to in-house OLAP development, where release would mean exposure to their most sensitive failure data).
IBM apparently uses OLAP on their hard drives because they don't quote their mean-time-between failures. I even called them and asked.
I've used their SAS programming language earlier in the year, and I'll tell you-- it's NO walk in the park. The syntax is worse than umm.. well.. I guess it's just the worse.
The only reasonable solution to this lad's problem would be to develop his own system.. It's the cheapest, probably the most reliable and, would be, by far, the most customizable.
I feel for him for having to do this kind of work though because I was driven mad by it. I still have a bad taste in my mouth from it.
My suggestion to him : hire a bunch of interns and have *them* do it.
I'm a home-grown American, and I can say that we don't *always* have that attitude. Our current measurement system isn't called English for our health.... >p>But the main reason we have that belief at all is because it seems everything foreign wants to blow up our government buildings ?? Maybe I'm just being paranoid...
And another thing-- If we converted to metric, I would cry for days because I wouldn't be able to coast on the expressways at 75 miles per hour.. It would be like 140 km/hr..
Or maybe I'm just being paranoid.
Your first theory cannot be correct. "it must be concentrated more densely near the sun and less densely farther away" violates one of the leading views on space and time-- that the universe looks and acts the same no matter where you are. The second theory-- If dark matter did exist, I seriously doubt it would have any effect on the space probe because A) It is too sparse B) If it weren't, we'd have found more by now And back in the day, people used the "Ether" theory to explain why light always travels the same speed and similar effects. It is my opinion that the apparent error in this probe's speed and acceleration are due to it's moving through space/time at 27,000 mph. This probably has a small effect locally, but over time it builds up and in effect, moves through time slower than you or I. Of course, that is all Einstein's general theory, and the NASA guys have already looked at that.. I would be really interested in seeing them post all of their equations and work so we can check for errors.