We plan on doing video of the learning process and posting it to our blog. In fact, right now we could do a video showing how the "ears" strip off the carrier wave. That is working now. I don't know how much you would get from seeing a voice on the oscilloscope and then seeing only the modulation wave after we hook up the filter though.
Until Dr. Jim finishes coding for VR, that's about all we can show. Give us about a month and we can start posting training episodes to our blog.
mallred said...
We have actually considered setting the robot in front of the TV and seeing how well it learns that way. Unfortunately, there is usually nothing intelligent on TV.
mallred said...
We plan on doing video of the learning process and posting it to our blog. In fact, right now we could do a video showing how the "ears" strip off the carrier wave. That is working now. I don't know how much you would get from seeing a voice on the oscilloscope and then seeing only the modulation wave after we hook up the filter though.
Until Dr. Jim finishes coding for VR, that's about all we can show. Give us about a month and we can start posting training episodes to our blog.
mallred
How about the pc / mainframe examples Dr. Jim refers to the ROBOT interview?· If that is the premise for the work being done for the Propeller, surely that would provide some insight as to the 'big picture'
mallred said...
We plan on doing video of the learning process and posting it to our blog. In fact, right now we could do a video showing how the "ears" strip off the carrier wave. That is working now. I don't know how much you would get from seeing a voice on the oscilloscope and then seeing only the modulation wave after we hook up the filter though.
(emphasis added)
Actually·Mark, I would be very interested in seeing exactly that. Please post the video!
Sure, I'll work with Dr. Jim on getting some video of the original voice on the oscilloscope, then hook up the filter, and let you see the difference when only the modulation wave is available.
"I don't know how much you would get from seeing a voice on the oscilloscope and then seeing only the modulation wave after we hook up the filter though." - you'd be surprised
Has anyone done a You-Tube search for "Dr. Jim's Design Corner" or "Machine Intelligence Conversation"? ... I'm just saying.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔ Beau Schwabe
IC Layout Engineer
Parallax, Inc.
> Has anyone done a You-Tube search for "Dr. Jim's Design Corner" or "Machine Intelligence Conversation"? ... I'm just saying.
Actually, I have watched all the Corner and Conversation clips.· But other than describing the general layout of the boards, io and file system (which while impressive are not unique in themselves, alternatives exist), there is really no 'conversation' or demonstration of machine intelligence; only Dr. Jim musing on the subject.
Referring to his interview in ROBOT, he suggests he has suceeded along these lines on pc / mainframe platforms... but I haven't seen or read anything about any kind of operational system.
I would like see a comparison with the great work of Dr. Stephen Falken on the WOPR since I guess that would be considered a Mainframe class machine...
> I would like see a comparison with the great work of Dr. Stephen Falken on the WOPR since I guess that would be considered a Mainframe class machine...
Agent420 said...
> I would like see a comparison with the great work of Dr. Stephen Falken on the WOPR since I guess that would be considered a Mainframe class machine...
Unfortunately, most of Dr. Jim's work in the past has been on imaging for the military. This is why we cannot show you the previous examples you are requesting.
Perhaps if you try the Freedom of Information Act, some of his material may be available.
Alternatively, Johnson and Johnson are running breast and prostate diagnosis machines with the core of Dr. Jim's image backscatter technology (a basis for his machine intelligence). You might also research that.
> Unfortunately, most of Dr. Jim's work in the past has been on imaging for the military. This is why we cannot show you the previous examples you are requesting.
DR. JIM: I have made some inroads in this. I have developed software that runs on PCs and some mainframes. I am now porting this over to a Parallax Propeller-based “hobby research super computer” setup. I cannot predetermine the internal interconnections it makes between its neuronal groups, and it can make mistakes and learn by correcting them.
ROBOT: Have you tested the synthetic brain in a verbal interaction?
DR. JIM: Yes; on a PC, but I really want it on the Parallax Propeller.
> Yes, a highly modified military PC specifically designed for his project. In fact, Dr. Jim had to develop the video card from scratch.
I fail to see the relationship between verbal interaction and a custom built video card, but if you say so...
Come on now, in fairness how can you say that in a magazine·interview and then backpeddle to claim it was all top secret, but you're porting it to the propeller?· When somebody says "I have developed software that runs on PCs", the inference is obviously a standard pc.· If he meant to say "...runs on specially modified military PC's designed for the project, which required building custom peripherals, including video cards, from scratch", they should just say that.·· It's that kind of doubletalk that gets you into trouble.
No. Dr. Jim says he was able to do that on a well known military project that is in use now. You would know it by name if I told you, but I'm not even supposed to know, technically. This robotic military device has been in use for years.
I have seen image technology that uses Dr. Jim's principles of machine intelligence learn to detect letters and groups of letters in such a way that it could "read". The demonstration I saw performed very well. This is the basis for my optimism.
But that's a neural network trick that was achieved years ago: Parallel Distributed Processing (1986 McClelland & Rumelhart. MIT Press ... the real MIT in Cambridge).
Joking aside, if past projects are classified that is fine and I can understand that. I think everyone would be a lot more receptive to your posts if you could start providing some real technical specifications or details on the current projects you are trying to sell. I would NOT expect those to be classified.
Why should we have to hunt down all this information to see if your projects are credible? Wouldn't it make sense for YOU to do that and post links on your site? I don't doubt that Dr. Jim is working on interesting stuff. However, I feel that many of the posts regarding the work has been OVERLY optimistic about things that haven't yet been demonstrated. I would personally love to see the research succeed but believe any plausible results are a bit further out than you realize. Hope I am wrong on that and the results turn up soon.
It's just that the robotics field is littered with so many false starts by people who completely underestimated the task at hand. Some in particular were hyped up by marketing departments before a working prototype was even complete. In one case (Genus robot) the prototype was never completed (at least that was what I learned from someone who worked on it) and the marketing pictures show the robot with a spreadsheet on the screen. It wasn't even a real screen, just a picture of one stuck on the shell to look like it since it wasn't working yet. They folded (as did many others) before finishing any products.
^ It's not really whether I believe it or not...· It's more a matter of describing your project as being mostly operational on a pc platform, which is certainly closer in proximity to the Propeller as far as optimistic·porting and realization is concerned, than discussing taking a classified military mainframe application and running it on a Propeller.· I see two entirely different degrees of association there.
Do I think it's possible?· Sure.· I've seen some crazy DARPA projects. Their 'Little Dog'·robot is amazing.· And I would hope that Dr Jim isn't doing all of this for naught.· But every time you start twisting what is going on, it gives me a headache while at the same time taking a bite out of your credibility.
Sure, I'll work with Dr. Jim on getting some video of the original voice on the oscilloscope, then hook up the filter, and let you see the difference when only the modulation wave is available.
No. Dr. Jim says he was able to do that on a well known military project that is in use now. You would know it by name if I told you, but I'm not even supposed to know, technically. This robotic military device has been in use for years.
I have seen image technology that uses Dr. Jim's principles of machine intelligence learn to detect letters and groups of letters in such a way that it could "read". The demonstration I saw performed very well. This is the basis for my optimism.
Thanks,
Mark
That's quite trivial. Recognising speech is much more difficult.
Leon
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Amateur radio callsign: G1HSM
Suzuki SV1000S motorcycle
But remember, Leon, we are not talking about regular OCR.· What I saw had machine intelligence behind it.· It is really imaging software, not OCR anyway.·
Dr. Jim's software was able to take the black and white photo you see above and make the below picture out of it.· OCR can't do that.· This was image processing imbued with machine intelligence software.· It even picked up on different wave patterns in the black and white picture and discovered a man-made underwater structure due to his technology.· I don't know that there is any other technology out there that can do this.
Anyway, your point is that OCR has been around a long time.· My point is that you can't compare apples and oranges.· OCR is very primative compared to this imaging technology designed by Dr. Jim.
No. Dr. Jim says he was able to do that on a well known military project that is in use now. You would know it by name if I told you, but I'm not even supposed to know, technically. This robotic military device has been in use for years.
I have seen image technology that uses Dr. Jim's principles of machine intelligence learn to detect letters and groups of letters in such a way that it could "read". The demonstration I saw performed very well. This is the basis for my optimism.
Thanks,
Mark
I wouldn't expect OCR to pick out arbitrary patterns from images like that in Robot Mag. It is intended for recognising printed characters and converting them into a text file.
Artificial neural networks "learn to detect letters and groups of letters". Has Dr Jim compared the performance of his system with that of an ANN?
Leon
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Amateur radio callsign: G1HSM
Suzuki SV1000S motorcycle
But it is not neural nets or anything from the past 30 years. Dr. Jim rejects all previous attempts because the results are so poor.
The key, from what I understand, is logic streams. He has his own definition for data that he calls logic streams. That is what is saved. That is what is interpreted as a concept, word, idea, etc. It is the uniqueness of each logic stream, and then the tying together of similar streams to form relationships in a cloud of neurons (memory). That is the way I understand it.
Again, the uniqueness of each logic stream (ECC-like code), in PROXIMITY (another key) with other like logic streams to create relationships among similar ideas.
Round - Ball - Sphere are all similar.
Oval - egg-shaped - eliptical - oblong. All similar, but ALSO related to the group above.
These words all have meaning and are synonyms, and then branch out to wider groups of like words or ideas. The proximity determines relationship and likeness. The further away something is stored, the less like the original nueron.
I have been thinking about Dr. Jim's approach, and this is how I understand it right now.
The Automatic Number Plate Recognition (ANPR) system used extensively in the UK uses ANN techniques, and is extremely accurate. There are many similar success stories.
How much better is Dr Jim's system for letter recognition than a standard ANN approach? Please provide some evidence.
Do you have a definition of a "logic stream" in the sense that Dr Jim is using it? I've seen it used in other contexts like fuzzy reasoning.
Leon
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Amateur radio callsign: G1HSM
Suzuki SV1000S motorcycle
Comments
Until Dr. Jim finishes coding for VR, that's about all we can show. Give us about a month and we can start posting training episodes to our blog.
mallred
Would it be possible to post a schematic for the audio prefilter? The photos alone don't reveal much that's useful.
Thanks,
-Phil
How about the pc / mainframe examples Dr. Jim refers to the ROBOT interview?· If that is the premise for the work being done for the Propeller, surely that would provide some insight as to the 'big picture'
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Mark
"I don't know how much you would get from seeing a voice on the oscilloscope and then seeing only the modulation wave after we hook up the filter though." - you'd be surprised
Has anyone done a You-Tube search for "Dr. Jim's Design Corner" or "Machine Intelligence Conversation"? ... I'm just saying.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Beau Schwabe
IC Layout Engineer
Parallax, Inc.
Actually, I have watched all the Corner and Conversation clips.· But other than describing the general layout of the boards, io and file system (which while impressive are not unique in themselves, alternatives exist), there is really no 'conversation' or demonstration of machine intelligence; only Dr. Jim musing on the subject.
Referring to his interview in ROBOT, he suggests he has suceeded along these lines on pc / mainframe platforms... but I haven't seen or read anything about any kind of operational system.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Hal 9000 vs WOPR Grudge Match
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
"Dave. I'm afraid I can't do that."
Perhaps if you try the Freedom of Information Act, some of his material may be available.
Alternatively, Johnson and Johnson are running breast and prostate diagnosis machines with the core of Dr. Jim's image backscatter technology (a basis for his machine intelligence). You might also research that.
mallred
DR. JIM: I have made some inroads in this. I have developed software that runs on PCs and some mainframes. I am now porting this over to a Parallax Propeller-based “hobby research super computer” setup. I cannot predetermine the internal interconnections it makes between its neuronal groups, and it can make mistakes and learn by correcting them.
ROBOT: Have you tested the synthetic brain in a verbal interaction?
DR. JIM: Yes; on a PC, but I really want it on the Parallax Propeller.
Military?
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
all you would have to do is copy from one platform to another. Data is data, the only difference is Dr. Jim's data is a wierd format (logic streams).
So, you could copy one robot to another and not have the learning curve. Dr. Jim has confirmed this will be possible.
Yes, a highly modified military PC specifically designed for his project. In fact, Dr. Jim had to develop the video card from scratch.
<edit out negative comment>
Post Edited (mallred) : 9/4/2009 2:48:37 AM GMT
Have you actually seen Dr Jim's prototype speech recognition and AI software working on the PC platform? How did it perform?
Leon
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Amateur radio callsign: G1HSM
Suzuki SV1000S motorcycle
I fail to see the relationship between verbal interaction and a custom built video card, but if you say so...
Come on now, in fairness how can you say that in a magazine·interview and then backpeddle to claim it was all top secret, but you're porting it to the propeller?· When somebody says "I have developed software that runs on PCs", the inference is obviously a standard pc.· If he meant to say "...runs on specially modified military PC's designed for the project, which required building custom peripherals, including video cards, from scratch", they should just say that.·· It's that kind of doubletalk that gets you into trouble.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Post Edited (Agent420) : 9/2/2009 7:55:30 PM GMT
No. Dr. Jim says he was able to do that on a well known military project that is in use now. You would know it by name if I told you, but I'm not even supposed to know, technically. This robotic military device has been in use for years.
I have seen image technology that uses Dr. Jim's principles of machine intelligence learn to detect letters and groups of letters in such a way that it could "read". The demonstration I saw performed very well. This is the basis for my optimism.
Thanks,
Mark
Post Edited (mallred) : 9/4/2009 2:49:13 AM GMT
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
--Steve
Propeller Tools
Why should we have to hunt down all this information to see if your projects are credible? Wouldn't it make sense for YOU to do that and post links on your site? I don't doubt that Dr. Jim is working on interesting stuff. However, I feel that many of the posts regarding the work has been OVERLY optimistic about things that haven't yet been demonstrated. I would personally love to see the research succeed but believe any plausible results are a bit further out than you realize. Hope I am wrong on that and the results turn up soon.
It's just that the robotics field is littered with so many false starts by people who completely underestimated the task at hand. Some in particular were hyped up by marketing departments before a working prototype was even complete. In one case (Genus robot) the prototype was never completed (at least that was what I learned from someone who worked on it) and the marketing pictures show the robot with a spreadsheet on the screen. It wasn't even a real screen, just a picture of one stuck on the shell to look like it since it wasn't working yet. They folded (as did many others) before finishing any products.
Robert
Do I think it's possible?· Sure.· I've seen some crazy DARPA projects. Their 'Little Dog'·robot is amazing.· And I would hope that Dr Jim isn't doing all of this for naught.· But every time you start twisting what is going on, it gives me a headache while at the same time taking a bite out of your credibility.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
That's quite trivial. Recognising speech is much more difficult.
Leon
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Amateur radio callsign: G1HSM
Suzuki SV1000S motorcycle
Leon
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Amateur radio callsign: G1HSM
Suzuki SV1000S motorcycle
Take a look at page 56 in the July issue of Robot Magazine http://www.machineinteltech.com/RobotMag7-09.pdf
Dr. Jim's software was able to take the black and white photo you see above and make the below picture out of it.· OCR can't do that.· This was image processing imbued with machine intelligence software.· It even picked up on different wave patterns in the black and white picture and discovered a man-made underwater structure due to his technology.· I don't know that there is any other technology out there that can do this.
Anyway, your point is that OCR has been around a long time.· My point is that you can't compare apples and oranges.· OCR is very primative compared to this imaging technology designed by Dr. Jim.
I wouldn't expect OCR to pick out arbitrary patterns from images like that in Robot Mag. It is intended for recognising printed characters and converting them into a text file.
Artificial neural networks "learn to detect letters and groups of letters". Has Dr Jim compared the performance of his system with that of an ANN?
Leon
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Amateur radio callsign: G1HSM
Suzuki SV1000S motorcycle
Post Edited (Leon) : 9/3/2009 12:19:32 AM GMT
The key, from what I understand, is logic streams. He has his own definition for data that he calls logic streams. That is what is saved. That is what is interpreted as a concept, word, idea, etc. It is the uniqueness of each logic stream, and then the tying together of similar streams to form relationships in a cloud of neurons (memory). That is the way I understand it.
Again, the uniqueness of each logic stream (ECC-like code), in PROXIMITY (another key) with other like logic streams to create relationships among similar ideas.
Round - Ball - Sphere are all similar.
Oval - egg-shaped - eliptical - oblong. All similar, but ALSO related to the group above.
These words all have meaning and are synonyms, and then branch out to wider groups of like words or ideas. The proximity determines relationship and likeness. The further away something is stored, the less like the original nueron.
I have been thinking about Dr. Jim's approach, and this is how I understand it right now.
How much better is Dr Jim's system for letter recognition than a standard ANN approach? Please provide some evidence.
Do you have a definition of a "logic stream" in the sense that Dr Jim is using it? I've seen it used in other contexts like fuzzy reasoning.
Leon
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Amateur radio callsign: G1HSM
Suzuki SV1000S motorcycle
Post Edited (Leon) : 9/3/2009 12:33:23 AM GMT