the thinker (neuron supercluster) by DRZOIDBERG

DRZOIDBERG

10 / 1

5th April

7th April

a neural network capable of influencing it's own biases, simulating learning to a certain extent. currently pretty useless, but it an interesting proof of concept either way.

brain learning machine network neural electricity electornics circuit neuron computer

Comments

• 

  klaotte003 11th April
  EnganK: pretty much anything can do that with a simple bias
• 

  EnganK 11th April
  I think you'd get a much better demonstration if this system had some simple goal. For example, two identical circuits, one trained to receive signal 1 and output signal 1, and the other trained to receive signal 1 and output signal 2. If you can't do even simple things like that, then your system can hardly be called a good proof of concept.
• 

  DRZOIDBERG 11th April
  This is in fact not the case. They can actually be quite simple and things like this are great examples of that. Designs this simple don't provide much in the sense of inputs and outputs while still maintaining acurracy and "training" effectiveness, but is a functional proof of concept nonetheless. The purpose of things like this is just to experiment with whats possible and they are not designed for practical use of any kind.
• 

  DRZOIDBERG 11th April
  it does take input and it does produce meaningful output. it's task is to send the correct output rather than the other. it is being called a supercluster as a way to describe several clusters working in unison that are connected together in a specif pattern to interact with eachother in specific ways. one of the biggest misconceptions surrounding neural systems is that they must always be extremely complex.
• 

  EnganK 10th April
  Also, you can't just randomly connect seven copies of the same circuit and call it a "supercluster"
• 

  EnganK 10th April
  There is no point in "training" and "biases" if your "neural network" does not accept any input and does not produce any meaningful output. Rewarding and punishing a neural network only makes sense when you have a specific task for it to perform. And since you don't have such a task, it's not a neural network, but simply a random signal generator.
• 

  DRZOIDBERG 10th April
  it's already a neural supercluster, so to theoretically make it into an AI you would need to hook several of them together with diffrent outputs giving them rewards and punishments and you would need to have their biases ajusted aswell, but thats really all there would be to it. it would take lots of orginizaition and it would be pretty difficult but in priciple the process would actually be pretty simple.
• 

  WaylonTheAmazing 9th April
  how can i turn it into basic ai?? then its nueral
• 

  DRZOIDBERG 9th April
  yes, all neural sytems are essentially randomizers at their core, but they have biases that are influenced by the reward/punishment sytem to make them more likley to go down the path that gives them a reward once it has received it enough times. This is whats called "learning" and all neural networks today are "trained" using a reward and punishment system that works similarly to this one.
• 

  EnganK 9th April
  Can the output be interpreted as real data? If not, it's simply a glorified randomizer.