Is it possibly to functionally transfer knowledge from a neural network to another?

Discussion in 'Intelligence & Machines' started by Buckaroo Banzai, Jan 3, 2018.

  1. someguy1 Registered Senior Member

    Messages:
    587
    This I find interesting. Link to more info?

    Are you claiming to have broken the Church-Turing thesis by constructing a computer that can compute more functions than a Turing machine can?

    I believe you that there is custom hardware to implement neural nets. But I don't believe you that this is a fundamentally new mode of computing. Such a result would have appeared in the news, as it would overthrow 80 years of thought in computer science.

    I'd be grateful for more context.
     
  2. Google AdSense Guest Advertisement



    to hide all adverts.
  3. billvon Valued Senior Member

    Messages:
    13,824
    Here's one: https://www.qualcomm.com/news/relea...ssing-engine-now-available-qualcomm-developer
    Not at all. It is just a device that can run neural networks more efficiently than a standard Harvard or von Neumann architecture.
    Not sure if you would call it a new _mode._ It is merely a new sort of hardware that allows more "direct" execution of neural networks. From the point of view of functionality you could simulate what the NPE does on a conventional computer; the NPE simply uses a new topology to do that much faster.
     
  4. Google AdSense Guest Advertisement



    to hide all adverts.
  5. someguy1 Registered Senior Member

    Messages:
    587
    Thanks for clarifying. Will check out the link.
     
  6. Google AdSense Guest Advertisement



    to hide all adverts.
  7. uhClem Registered Member

    Messages:
    16
    This video might be of interest to some here. It is not specifically about ANNs. Instead it is about the more general subject of AI. It is a presentation on DARPAtv by John Launchbury who is Director of the Information Innovation Office of DARPA. The video not too long at 16 minutes.

     
  8. Write4U Valued Senior Member

    Messages:
    6,912
    I believe that's what I said.
    The human brain is capable of abstract thought from pure logic. It does not need to follow strict algorithms. I believe we call it imagination.
    That is the point I am making. But the human brain's algorithm is not a fixed equation, but has a flexible organizational ability.
    Check this out.
    https://www.ted.com/talks/anil_seth_how_your_brain_hallucinates_your_conscious_reality
    Correct. Lemurs and many other animals are employing naturally evolved hardwired mathematical abilities, already allowing for problem solving and making some very sophisticated decisions. IOW, creating new synapses which makes the brain an evolving and learning organ with ever greater capabilities. Don't forget, man-made symbolic mathematical algorithms are individually fixed algorithms. OTOH, natural brain processes are learned sensory experiences, which creates a much different electro-chemical computational process than from specific programs in computers.
    IMO, this is what AI tries to achieve, a artificial brain capable of learning and append it's program to include variables. We already use the term "intuitive computers" which is a man-made evolutionary step above "fixed program" computers.
    NP, I've done it myself.

    But it is revealing that the computer did not warn you of a possible conflict in assigning authorship of your quotes, whereas James and I immediately noticed the inconsistency......

    Please Register or Log in to view the hidden image!

     
  9. someguy1 Registered Senior Member

    Messages:
    587
    I believe you and I are on the same side of the issue. I don't believe that minds are algorithms. Not sure why my words were misunderstood but not the first time it's happened.

    Pure logic is algorithmic. I agree with your main point that the mind is not an algorithm. I think you are muddling your point to toss in "pure logic" here. What is pure logic? First order, second order, modal, etc. Better to drop that point from your argument IMO since it's easily falsified or misunderstood.


    If it's not fixed it's not an algorithm. You're muddling your point I believe. An algorithm by definition is a description of a program. It's fixed on day one and then implemented in hardware (digital, person with pencil and paper, etc.)
     
    Last edited: Jan 10, 2018
  10. Write4U Valued Senior Member

    Messages:
    6,912
    I was not trying to argue against your perspective, just trying to eaxpound a little on it.
    I understand what you are saying and I agree to a certain extend.
    IMO, if an algorithm is a form of logic, the logic forms algorithms. Apparently the ability to do fundamental subconscious mathematical calculations, does suggest a mind which forms algorithms (albeit very simlple as algorithms go)

    But it is the apparent ability of the brain to be able to create algorithms on the fly at some subconscious molecular level, such as appending the forming image creating or using a "virgin" group of synapses to be able to extend the calculation abilities or "consult" possibly previously related experiences into continuous process of abstract thought (imagination).

    But even if the information is processed by electro chemical interactions, those interactions would be subject to the mathematical laws governing such interactions and that includes algorithmics.

    As Anil Seth proposed, the brain is a biological predictive engine, because it needs to make sense (best guess) of the original object out there, it's not physically inside the brain. The brain receives its information indirectly as translated by the sensory system, such as vision.
    Then it needs to process this stream of incoming information into a coherent whole picture of what all the different senses tell is out there.
     
  11. kx000 Valued Senior Member

    Messages:
    4,378
    Idk if telepathy is reality or not, but it is a human right to believe.
     
  12. Write4U Valued Senior Member

    Messages:
    6,912
  13. iceaura Valued Senior Member

    Messages:
    26,294
    There are some complexities you seem to be overlooking - especially regarding the direction of motion of currents, and the like. What exactly do you mean by "freeze the CPU"?
    Whether or not the human brain is a digital computer is not relevant - it is a physical object, and it can be in principle "frozen", its every detail of physical status duplicated, as with a CPU.
    The Go playing computer I mentioned presents the same kinds of "mysteries" to analysts as human Go players present. And knowing all about how its hardware and software are functioning at any given moment does not tell you what you need to know, to explain its play of the game.
    That would not help you very much in analyzing the play of the game, with the 9-dan Go playing computer.
    You would not be able to tell, from that, why it was making the moves it was making.
     
  14. Michael 345 In Aust : found it :) Valued Senior Member

    Messages:
    5,651
    I see a problem in freezing the activities of the brain (ie the electrical impulses traveling alone the neurones and the chemical reactions between the axions) and then building a exact copy of the brain to download the frozen program into (what would the interface look like?) and then booting the system back up

    Apart from all the above you have frozen the program at a particular instant. Which gives no indication of what the program is going to do next

    Please Register or Log in to view the hidden image!

     
  15. Write4U Valued Senior Member

    Messages:
    6,912
    Brian Greene did a presentation on NOVA about entanglement. In theory it might be possible to create an entangled duplicate of a person, then via entanglement reversal create a duplicate original.
    Not only that, but your duplicate might be thousands of miles from the original. A form of "entanglement" travel.
     
  16. Michael 345 In Aust : found it :) Valued Senior Member

    Messages:
    5,651
    Don't know or understand enough about entanglement to make a serious reply

    I think my point still stands
    Freeze Number 1 brain (disregard the time taken to do the freezing) and download load the conditions of brain at that moment (disregard the time taken to download)

    Upload to clean slate Number 2 brain with identical physical construct (disregard upload time)

    Reboot both same time

    The synch will not last even 1 second

    From the moment of reboot they will diverge

    If you are posting the brains will remain entangled I still do not think that will translate into identical thoughts

    Please Register or Log in to view the hidden image!

     
  17. Write4U Valued Senior Member

    Messages:
    6,912
    Can "thoughts" (electrical/chemical) values be "entangled"? I guess, theoretically one could copy every neuron and cell through entanglement, but the thought process itself?
     
    Last edited: Jan 14, 2018
  18. Michael 345 In Aust : found it :) Valued Senior Member

    Messages:
    5,651
    I would consider the thought PROCESS to be the collective electrical signals flowing along the neurones and being transferred by chemical reactions across synapses. That for me is the PROCESS

    What is the interface in the brain with turns such a process into voices and images within the brain - no idea

    Duplicate the process - no problem

    Duplicate the thoughts - problem

    It might might might just be possible to construct a human brain (perhaps by using a animal brains - connected both in series and parallel formats)

    A EEG would indicate its working however extracting the thoughts would be like tuning a TV with you having no ability (interface or equipment to obtain a meaningful translation of the process)

    Yes the EEG shows a PROCESS is in operation by giving you ink squiggles on graph paper

    But you can obtain a credible EEG from attaching electrodes to a jelly. Credible enough to cause a doctor to think the jelly is alive and not to turn off life support

    Please Register or Log in to view the hidden image!

     
    Write4U likes this.
  19. someguy1 Registered Senior Member

    Messages:
    587
    Very interesting point. Digital circuitry inputs electricity in analog form and, through the miracle of electronics, outputs a digital signal. So that although at a deep level a computer consists of electrons flowing and sloshing around; but at the level of the computer, it's crisp 1's and 0's.

    The way it's done is with a system clock. The clock is a specialized circuit that lets out one tick every tiny time interval. When you hear about the cpu in your computer being rated at 3.0GHz, that means its clock sends out a little tick 3 billion times a second.

    Now every other circuit in the computer is keyed to the clock tick. At each tick, a low-level machine instruction executes in the cpu; causing the cpu to flip certain bits in its registers and/or transfer bit patterns from its registers to or from memory.

    After the completion of each instruction, we could take a snapshot of the state of the cpu and memory. To freeze the cpu means to just take that snapshot. Hardware engineers who design cpus most likely have a special circuit that does literally freeze a cpu. A computation is just a sequence of snapshots starting from some initial state.

    That's exactly what digital computers are, and what they do.

    Oh but this is NOT true; or at the very least, it's very far from being known. We have no theory of how the brain works from one instant to the next. There's no proof it's digital at all and in my opinion it isn't.

    You are making a metaphysical assumption about the nature of the world. Your assumption goes far beyond anything known to contemporary physics.

    If you would agree that you are making an assumption, I'd grant you the point for sake of conversation. But if you claim this assumption as a scientific truth, you are misinformed about that. In the physical world we do not have a concept of an "instant." We only have instants in our mathematical models. In the real, physical world ... we simply do not know.


    This may be true, although in fact even the most advanced neural net runs on conventional hardware and its operation can be described as a sequence of states of the cpu and memory. Sure it's awesome how Alpha Zero works but that does not imply anything about metaphysics!

    Yes ok. I agree with your observation, but I don't see this as being as meaningful as you do.
     
    Last edited: Jan 15, 2018
  20. someguy1 Registered Senior Member

    Messages:
    587
    It's amazing what the brain does. But there's no scientific consensus on whether algorithms are one of its core operating principles.

    And "creating algorithms on the fly" is poetic but does not make any sense within computer science. You are positing some brain function that, given a situation, outputs an algorithm.

    It sounds really cool but it doesn't mean anything. There is no such concept in the scientific study of algorithms. Nor any known such brain function in neuroscience.
     
  21. Write4U Valued Senior Member

    Messages:
    6,912
    If one wanted to map the entire sequence accurately, would we not have to take 3 billion snapshots of the neural network every second?
     
  22. Write4U Valued Senior Member

    Messages:
    6,912
    True, the brain works by recognizing patterns according to association with previously compartmentalized knowledge.

    Algorithms (as we understand and use them) are not required (hence the poetry), but we have a peculiar inate ability to evaluate the information from our 5 senses which provide 5 different sensory experiences, which internally seem to be be directly connected to each, for comparison and possible association. This ability allows us to analyze a phenomenon from different perspectives, which eventually allow us to understand why and how the patterns in the observed phenomena occur and from this knowledge we have created the algorithms wich drive our computers. Algorithms are approximations of what and how we can process information in an binary electronic device.

    This subconscious function produces "intuitive thinking" or, as Anil Seth proposes, it allows the brain to make "best guesses", which then can be tested and translates into symbolic language such as mathematical equations and computer algorithms.
     
  23. someguy1 Registered Senior Member

    Messages:
    587
    No. If we like we can take the initial state and a written copy of the algorithm -- the program -- and work it out with pencil and paper and take ten or fifteen minutes per tick.

    We can follow the exact same instructions the computer does.

    We would perform the exact same computation. Slowly to be sure. But that's one of the key properties of algorithms. They compute exactly the same thing independently of the physical nature of the hardware.

    In fact this is exactly what programmers do when they are debugging. They step through the program line by line, writing down the contents of the memory locations they care about, till they see where they made their programming error.

    It does not matter how fast a computation runs or what is the nature of its hardware.

    A given algorithm always does exactly the same thing.

    This is IMO an argument against mind being some kind of property or output of a computation. If this were true, then a person sitting with a big box of pencils and a big stack of paper and a copy of the computer program could start at the first line and execute that program line by line and thereby implement a mind. And where exactly would that mind be? When does the mind come into existence?

    If mind is any kind of byproduct or output of a computation that depends on the speed of execution, then mind, whatever it is, is not computational.

    It's a defining property of an algorithm that whatever it does, it does it the same whether it goes fast or slow.
     
    Last edited: Jan 15, 2018

Share This Page