@wenhurley In this case I just used an equation which looked much like a mesa you might see in the midwest. In Excel this would need to run overnight to get close to the answer. Excel's overhead requirements make it impractical for real use but it's reasonable demo. IRL I'd do this in VB or something.
completely uselsss. I used to teach NN and AI 20 years ago, and I would never have shown something without trying to explain it.
the kid who did that probably has no clue of what he was doing, but if you push him a bit, maybe he'll look up something online and could give some "valid" explanations of this. :-)
@robextra0 I find it hard to believe you've ever taught anything. Community college? What? When you were 18? You sound like a 14 year old troll not an adult. Looking at the comments you leave other people, I can see you're just a whiner. How pathetic. I'm having discussions with other interested people here but you come along a complain about something you probably don't understand. You probably live in your mom's basement and spend your time on Xbox and porn. Screw you.
@cybercab: I am not the first one to say that here, so again: you are showing some nice 3D graphs without explaining what it is. does this show the convergence of the coefficients of a multilayer network during its learning phase or what?
BTW, your post below saying "it's a waste of your time" to explain what it is here shows how pompous and arrogant you are, hoping to impress without trying to explain. why don't you go LEARN something, then make yourself useful and TEACH it back?
@cybercab actually looking at the video again and the posts below, it's not just about convergence. Anyway, a couple of sentences upfront could have made it worthwhile and saved you some aggravation over the years.
@robextra0 It has worked just fine for me. I show it to people who want to see what I'm working on. It's not here for the use of people like you. You don't matter.
@Houshalter This is the visual representation of a NN I built in Excel. I posted it so I could explain to people at work what I am doing. You're welcome to come by my desk or check out Gradient Descent NNs on Wikipedia. It's a waste of my time to explain it again here. Cools stuff though.
More or less. The number of weights doesn't change. It's simply adapting. Another way to think about it is that every point on the plane shifts slightly. Then all data is run through the new plane to see if it is a better fit. If not it reverts. If better then that becomes the new baseline. It's not perfect but it's pretty good in most situations.
@cybercab Yes that makes more sense than what i sugested. I thought you were trying to show the convergence of the algorythm on certain values for each weight. Where the weights were numbered in the 2d plane of the base and their value was the height, but then the nn would have 25 layers.... too many.
Yeah, I don't take harassment well. I've seen nerds argue before but that last guy was insulting and incorrect. Maybe he'll figure this stuff out one day. It's pretty fun stuff.
I'm using a neural network with a gradient descent approach. It's a form of artificial intelligence. This data was just a sample but nothing real. The use of this is that it will find the relationship between 2 or more things. Then generate the equation you can use to be more efficient. In the market you might want to know how to invest when Oil=$x, steel=$y, labor=$z. Too complicated for people to understand but this approach will figure it out. Works on many problems,
The neural network is testing the solution against all known data from the past. Then slightly alters its guess as to what the right answer is. Over time it "feels" its way to the right answer. It's an amazing technique using artificial intellegence invented in 1986.
Its more accurate to say it was "perfected" in 1986. The mathematics was sort of derived independently by other people namely Yan LeCun and Paul Werbos, but it was for areas different than neural networks, but needing nonlinear regression which is what this is doing.
BUT you should see the video Next Generation Neural Networks where they discuss shortcomings of "backpropagation" learning and now focus on "generative models" where the network learns to mimic, rather than classify the dataset.
AI is old. Turing used is in WWII. Neural Networks are generally accepted to have been invented in 1986. It's true backpropagation was popular in 1974 but that's really not the same thing.
Well we're all very impressed with your knowledge but I don't think your facts are quite right. You have some interesting videos too. Do you have permission to post all that stuff or are you just taking your own approach to that as well?
How is Karl Rove on topic? Do you have a PhD is this stuff? Do you build neural networks? This NN uses gradient descent approach which was a big deal in 1986. My associate was taught by the guy who developed the algorithm when he got his PhD at Carnegie Mellon.
So do you do this stuff or just read a lot? Turing's big paper came out in 1948. Since then there have been NNs in many forms. So do you actually use them for anything? I use some is business but mostly I just see people misuse them.
What is the problem this guy is trying to solve in Excel? looks like the training of the NN, but by the movements the NN has reached nothing
wenhurley 1 year ago
@wenhurley In this case I just used an equation which looked much like a mesa you might see in the midwest. In Excel this would need to run overnight to get close to the answer. Excel's overhead requirements make it impractical for real use but it's reasonable demo. IRL I'd do this in VB or something.
cybercab 1 year ago
@cybercab If you know Java or C#, google "heaton research" and you could find something useful for your new developments
wenhurley 1 year ago
This has been flagged as spam show
@cybercab If you know Java or C#, google "heaton research" and you could find something useful for your new developments
wenhurley 1 year ago
completely uselsss. I used to teach NN and AI 20 years ago, and I would never have shown something without trying to explain it.
the kid who did that probably has no clue of what he was doing, but if you push him a bit, maybe he'll look up something online and could give some "valid" explanations of this. :-)
robextra0 1 year ago 2
@robextra0 I find it hard to believe you've ever taught anything. Community college? What? When you were 18? You sound like a 14 year old troll not an adult. Looking at the comments you leave other people, I can see you're just a whiner. How pathetic. I'm having discussions with other interested people here but you come along a complain about something you probably don't understand. You probably live in your mom's basement and spend your time on Xbox and porn. Screw you.
cybercab 1 year ago
@cybercab: I am not the first one to say that here, so again: you are showing some nice 3D graphs without explaining what it is. does this show the convergence of the coefficients of a multilayer network during its learning phase or what?
BTW, your post below saying "it's a waste of your time" to explain what it is here shows how pompous and arrogant you are, hoping to impress without trying to explain. why don't you go LEARN something, then make yourself useful and TEACH it back?
robextra0 1 year ago 2
@cybercab actually looking at the video again and the posts below, it's not just about convergence. Anyway, a couple of sentences upfront could have made it worthwhile and saved you some aggravation over the years.
robextra0 1 year ago
@robextra0 It has worked just fine for me. I show it to people who want to see what I'm working on. It's not here for the use of people like you. You don't matter.
cybercab 1 year ago
What does this have to do with NNs? All I saw was a series of random graphs. It'd be nice if you'd explain whats going on in the description.
Houshalter 1 year ago
@Houshalter This is the visual representation of a NN I built in Excel. I posted it so I could explain to people at work what I am doing. You're welcome to come by my desk or check out Gradient Descent NNs on Wikipedia. It's a waste of my time to explain it again here. Cools stuff though.
cybercab 1 year ago
Are we watching the output data? or you are showing the weights change through the iterations in a 3d graph?
goliathlup1 2 years ago
More or less. The number of weights doesn't change. It's simply adapting. Another way to think about it is that every point on the plane shifts slightly. Then all data is run through the new plane to see if it is a better fit. If not it reverts. If better then that becomes the new baseline. It's not perfect but it's pretty good in most situations.
cybercab 1 year ago
@goliathlup1 We're looking at the current guess of the final answer.
cybercab 1 year ago
@cybercab Yes that makes more sense than what i sugested. I thought you were trying to show the convergence of the algorythm on certain values for each weight. Where the weights were numbered in the 2d plane of the base and their value was the height, but then the nn would have 25 layers.... too many.
goliathlup1 1 year ago
Jeez dude... somebody doesn't behave like a 37yo
OriginalAtomicSheep 2 years ago
Yeah, I don't take harassment well. I've seen nerds argue before but that last guy was insulting and incorrect. Maybe he'll figure this stuff out one day. It's pretty fun stuff.
cybercab 2 years ago
Can you fill in the blanks with this video? What problem is it trying to solve? What method is it using to solve it?
hypoxide 3 years ago
I'm using a neural network with a gradient descent approach. It's a form of artificial intelligence. This data was just a sample but nothing real. The use of this is that it will find the relationship between 2 or more things. Then generate the equation you can use to be more efficient. In the market you might want to know how to invest when Oil=$x, steel=$y, labor=$z. Too complicated for people to understand but this approach will figure it out. Works on many problems,
cybercab 3 years ago
uMMm like O-M-G this is so stupid HAhahha P
JJAKE543 3 years ago
Actually it's monumentally significant. Artificial intelligence will one day help the state care for mentally challenged people such as yourself.
cybercab 3 years ago
hi, are you intrested in working with me on an excel NN project? if so email me,,, mehul1977 at hotmail . co . uk
mehul1977 3 years ago
???????????????????
telepathicoincidence 4 years ago
The neural network is testing the solution against all known data from the past. Then slightly alters its guess as to what the right answer is. Over time it "feels" its way to the right answer. It's an amazing technique using artificial intellegence invented in 1986.
cybercab 4 years ago
Its more accurate to say it was "perfected" in 1986. The mathematics was sort of derived independently by other people namely Yan LeCun and Paul Werbos, but it was for areas different than neural networks, but needing nonlinear regression which is what this is doing.
BUT you should see the video Next Generation Neural Networks where they discuss shortcomings of "backpropagation" learning and now focus on "generative models" where the network learns to mimic, rather than classify the dataset.
nightowl8936 3 years ago
AI is old. Turing used is in WWII. Neural Networks are generally accepted to have been invented in 1986. It's true backpropagation was popular in 1974 but that's really not the same thing.
cybercab 3 years ago
"Neural Networks are generally accepted to have been invented in 1986."
No. You are incorrect. Google around and see that I am right.
1943 - McCulloch and Pitts - networks of boolean logic gates - the earliest neural network based on known neurophysiology at the time.
1957 - Rosenblatt - Perceptron - linear classifier
1960s - Hoff et. al. - Delta Rule
1986 - Hinton et. al. generalized delta rule + multilayer perceptron = backpropagation.
Trust me I know what I am talking about.
nightowl8936 3 years ago
Well we're all very impressed with your knowledge but I don't think your facts are quite right. You have some interesting videos too. Do you have permission to post all that stuff or are you just taking your own approach to that as well?
cybercab 3 years ago
"I don't think your facts are quite right."
I can assure you I know the history of back-propagation neural networks backwards and forwards.
I been studying this stuff since the 90's, and I am beginning to explore neural network chips with VHDL. So prove me wrong if you can.
"Do you have permission to post all that stuff or are you just taking your own approach to that as well?"
Are you a Republican or something? Spare me your Karl Rove tactic. I know them all. Just stay on topic.
nightowl8936 3 years ago
How is Karl Rove on topic? Do you have a PhD is this stuff? Do you build neural networks? This NN uses gradient descent approach which was a big deal in 1986. My associate was taught by the guy who developed the algorithm when he got his PhD at Carnegie Mellon.
So do you do this stuff or just read a lot? Turing's big paper came out in 1948. Since then there have been NNs in many forms. So do you actually use them for anything? I use some is business but mostly I just see people misuse them.
cybercab 3 years ago
"Karl Rove on topic?"
Off on tangent - distract from main issue.
"Phd in this stuff?"
Master's level courses /use of open-source simulators.
"Do you build NNs?"
In software, working towards embedded systems for my own company
nightowl8936 3 years ago
"gradient descent ... big deal in 86"
The key was fixing "credit assignment problem" for multiple layers / weights, see "Perceptrons"
"Turing's paper" - Turing test = Symbolic AI
I use NNs as function approx for engineering.
nightowl8936 3 years ago
hi, are you intrested in working with me on an excel NN project? if so email me,,, mehul1977 at hotmail . co . uk
mehul1977 3 years ago