Basics of neural networks and connectionist architectures

Discussion in 'Intelligence & Machines' started by Magical Realist, Aug 25, 2013.

  1. Magical Realist Valued Senior Member

  2. Google AdSense Guest Advertisement

    to hide all adverts.
  3. Billy T Use Sugar Cane Alcohol car Fuel Valued Senior Member

    Yes that is a useful link. It even tries a little to call the computer a connection machine (instead of a neural network) - A battle I lost decades ago. It is best to skip all the first part about the brain and jump right in after this diagram:

    Please Register or Log in to view the hidden image!

    The first part mis leads: Connection machines don't use any thing like neural discharge pulses. (Another reason why they should not be called "neural networks.") They are analogue devices but most often emulated in digital machines at least before you make many say to be the intelligence in a torpedo helping it decide how to track ship it "wants" to sink. Especially how far below the ship it should explode, if ship is big - that is best for big ships instead of exploding against the hull as most ill informed people think from watching too may WWII films. (Let the middle of the ship fall into the big gas bubble while water supports both ends. Ships are not strong against that type of bending - even just improper loading of an oil taker can break it apart. Tanker MUST remain nearly level as it sinks deeper into the water. - If not, one end will crack off.)

    BTW, it has been proven a few decades ago back in the early era of connection machines, that adding another "intermediate" layer does not produce a more capable machine. (Or more accurately stated: there is no problem a four (or more) layer machine can solve that a properly designed 3 layer machine can not.)

    Yes it is amazing how well these machines can do on complex yes no decisions - better than humans for example in deciding whether or not to grant bank loan based on the data only (Machine does not see female flirt with the bank clerk, etc.)

    In fact computational machines of some sort may take over - old "science fiction" story that may soon be science reality.. Read more here: At least that is how intelligence will migrate from Earth to some distant planet.
  4. Google AdSense Guest Advertisement

    to hide all adverts.
  5. TruthSeeker Fancy Virtual Reality Monkey Valued Senior Member

    We always only have 2 choices at a time, though they tend to come in bundles, creating the illusion that we have multiple choices. Regardless of the network, this will always hold true. Simplifying our perception of the networks and the systems will lead us to a deeper understanding of the choices available at any point within the system and, ultimately, enable us to disintermediate the systems.

    I would say it can be a bit deceiving to understand these systems as inputs, "hidden" and outputs. In reality, there is a previous choice which had an outcome (the input), a current choice and the outcome of the current choice (the output), which leads to the future choices. The "hidden" part can be better understood as "processing", which is essentially a rule that dictates how a choice between 2 alternatives is to be made. Ultimately, we perceive those systems as a collection of inputs, processes and outputs, and there can be many steps to a process before an "input" is fully transformed into an "output".
  6. Google AdSense Guest Advertisement

    to hide all adverts.
  7. Billy T Use Sugar Cane Alcohol car Fuel Valued Senior Member

    Hard to tell if you speak of connection computers or of brain processes. People certainly have more than one choice at a time - for example which of say 8 horses to bet on as the race winner.

    For the brain, it may be true, but I doubt it always is, that the "input" to some stage is the output from some prior stage of neural processing, but certainly could be from many different possible actions being considered. For example which of 16 different flavor of ice-creams, the store is selling, is the one you want. Most man-made connection computer do NOT get their "inputs" for the outputs of other computers. For example if predicting rain or no rain tomorrow in NYC, there could be several dozen of factual inputs both from historical records and from current weather data.

    SUMMARY: Your post seems very confused and unclear to me. If you have some point, please state it more clearly.
  8. TruthSeeker Fancy Virtual Reality Monkey Valued Senior Member


    No, there are only 2 choices at a time in that example, though it is a collection of 2 choices. Here are they:
    1) A choice between horse 1 or any of the others.
    2) A choice between horse 2 or any of the others.
    3) A choice between horse 3 or any of the others.
    4) A choice between horse 4 or any of the others.
    5) A choice between horse 5 or any of the others.
    6) A choice between horse 6 or any of the others.
    7) A choice between horse 7 or any of the others.
    8) A choice between horse 8 or any of the others.

    The choice is complex because you have to be constantly shifting your attention from one horse to another, creating this illusion that we are choosing between 8 horses. You always have to choose between "A" and "Not A". The difference is that "A" is fairly defined while "Not A" is not as defined.

    Inputs and outputs can come in various forms. There are inputs and outputs that are external to the brain and inputs and outputs that are internal. This is true to any system or collection of systems. You could visualize it this way:

    External Input -> [process -> output/input -> process -> output/input... -> process] -> External Output
  9. Billy T Use Sugar Cane Alcohol car Fuel Valued Senior Member

    No still false and that stilted logic only applies to selection between discrete choices, not for example when I am cutting up one large steak to give to wife and kids each a piece. There the choice as to where to cut is a continuous variable, as are many if not most choices, like when to leave a party, how much beer from the keg to put in your empty glass, etc. Your "A or not A" gets very stilted / artificial when the choice is some point in a continuum.
  10. TruthSeeker Fancy Virtual Reality Monkey Valued Senior Member

    There is certainly various degrees within a choice, but these choices can also be framed as "do I cut up this size of steak or another size?" , and "Do I leave the party right now or another time?", "do I put this amount of beer or should I put another amount (more or less)?"

    The decision always seems to happen between your final choice and all the other possible ones.

    It appears to me that everything can be reduced to a dualism, though as you point out, there is an element of continuous spectra as well...

Share This Page