Page 2 of 2 FirstFirst 12
Results 11 to 20 of 20

Thread: How do they make more accurate machines, if it takes machines to make machines?

  1. #11
    Insider PBSteve's Avatar
    Join Date
    May 2012
    Location
    Los Angeles
    Posts
    2,881
    That said, using a neural net might be an interesting way to develop a new control scheme.

    Edit: I'm reading the paper and I think that's what Ryan was talking about, although it was unclear in the way he worded it.
    Last edited by PBSteve; 01-27-2017 at 12:59 PM.
    "He died on that hill even though no one was attacking"
    I work for the company building the Paragon...once we figure out a name

  2. #12
    no, i think its more like, you use a neural network when you don't know the fundamental governing equations. you use the network to derive its own governing equations given a certain set of input data. the the netowrk not only has to use the governing equations, it also has to figure out what they are.

    in most cases where modeling is too complex or too tough for engineering applications, we know the governing equations. and the governing equations and there application is in and of itself, too complex for computational solutions to be generated in a reasonable time. and thats when you know and understand the system, and why its behaving as it is. if you don't know that, like a neural network wont, then you have added a HUGE mess of computation that your network has to do. it has to figure out what the problem is that it needs to solve. whereas, we already know how to solve the problem.
    social conservatism: the mortal fear that someone, somewhere, might be having fun.

  3. #13
    Insider
    Join Date
    May 2012
    Posts
    1,158
    Quote Originally Posted by PBSteve View Post
    I suspect that's true for a huge number of applications.

    Neural models are still kludgy at best, and even when they get better they won't always be appropriate. If you're working on a system with linear functions (like temperature), and just a few scalar ins and outs, there's no reason to introduce a neural network - your computer is doing nothing more than addition and multiplication, something a binary calculator is exceedingly good at. A huge portion of science engineering fits into this category, of simple, low-dimension linear relationships.

    It'd be like using a quantum computer to run MS Word. You don't need a neural net to run a thermostat, in fact it's probably a bad idea.
    I agree with all of this, with the pithy addendum that it is by their very kludginess that nets are useful at all. A lot of the criticisms of them as a toy have been fixed by Nvidia bothering to write cuDNN.

    My assumption in the original post is that Gordon actually has an interesting problem on his hands in the remaining noise that he can't dial out by his pre-cut and compensate approach. (A feedback mechanism superior to feedforward mechanisms which tend to have divergence problems.)

    Something like taking all of your input data and then looking at accuracy error in an REML fit model analysis could be useful to try to poke around if anything you're ALREADY MEASURING could be assignable in further error. I'm pretty sure linearity would be baked into that and it's computationally very cheap. The attractive part about doing it with online machine learning is you wouldn't have to fuck around with orthogonal DOE-by-hand nonsense and can just gradient descend you way to a local optimum.
    "So you've done this before?"
    "Oh, hell no. But I think it's gonna work."

  4. #14
    Insider
    Join Date
    May 2012
    Posts
    1,158
    Quote Originally Posted by cockerpunk View Post
    no, i think its more like, you use a neural network when you don't know the fundamental governing equations. you use the network to derive its own governing equations given a certain set of input data. the the netowrk not only has to use the governing equations, it also has to figure out what they are.

    in most cases where modeling is too complex or too tough for engineering applications, we know the governing equations. and the governing equations and there application is in and of itself, too complex for computational solutions to be generated in a reasonable time. and thats when you know and understand the system, and why its behaving as it is. if you don't know that, like a neural network wont, then you have added a HUGE mess of computation that your network has to do. it has to figure out what the problem is that it needs to solve. whereas, we already know how to solve the problem.
    Technical quibbles aside, I think this is a good way of thinking about it - it was unclear from your original post that you "already knew how to solve the problem".
    "So you've done this before?"
    "Oh, hell no. But I think it's gonna work."

  5. #15
    Insider PBSteve's Avatar
    Join Date
    May 2012
    Location
    Los Angeles
    Posts
    2,881
    Ah, got it. I think I'm caught up now.
    "He died on that hill even though no one was attacking"
    I work for the company building the Paragon...once we figure out a name

  6. #16
    Quote Originally Posted by Lurker27 View Post
    Technical quibbles aside, I think this is a good way of thinking about it - it was unclear from your original post that you "already knew how to solve the problem".
    modeling, meshing, boundary conditioning, and initializing an entire diamond turning machine, and then modeling every interaction that shows how such a machine changes size based on fractional degrees of temperature gradients, and the performance of the cooling systems ... i mean, its all technically possible. its a solvable problem.

    its just computationally so complex it might as well be unsolvable.
    Last edited by cockerpunk; 01-27-2017 at 01:28 PM.
    social conservatism: the mortal fear that someone, somewhere, might be having fun.

  7. #17
    Insider PBSteve's Avatar
    Join Date
    May 2012
    Location
    Los Angeles
    Posts
    2,881
    Quote Originally Posted by cockerpunk View Post
    its just computationally so complex it might as well be unsolvable.
    Kind of sounds like an interesting problem to apply a neural net to.
    "He died on that hill even though no one was attacking"
    I work for the company building the Paragon...once we figure out a name

  8. #18
    Quote Originally Posted by PBSteve View Post
    Kind of sounds like an interesting problem to apply a neural net to.
    my bet is that my machine does not have fundamental governing equations that are not simply first principles equations piled on top of one another.
    social conservatism: the mortal fear that someone, somewhere, might be having fun.

  9. #19
    Junior Member
    Join Date
    Dec 2017
    Posts
    1
    Quote Originally Posted by cockerpunk View Post
    modeling, meshing, boundary conditioning, and initializing an entire diamond turning machine, and then modeling every interaction that shows how such a machine changes size based on fractional degrees of temperature gradients, and the performance of the cooling systems ... i mean, its all technically possible. its a solvable problem.

    its just computationally so complex it might as well be unsolvable.
    Do you think computational power of the future will be able to solve problems like that?

  10. #20
    Quote Originally Posted by FernandoSel View Post
    Do you think computational power of the future will be able to solve problems like that?
    maybe.

    i have a hard time believing that just measuring it and controlling it isn't a cheaper/easier/more accurate method to solve the problem.

    #boundryconditionsareabitch
    social conservatism: the mortal fear that someone, somewhere, might be having fun.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •