Vanilla 1.1.9 is a product of Lussumo. More Information: Documentation, Community Support.

  1.  
    http://www.kdnuggets.com/2014/12/geoffrey-hinton-talks-deep-learning-google-everything.html

    I am still staggering through Tim Urban's two part article on waitbutwhy. It's very good. I'm worried whether I should be worried about military shits getting hold of Deep Learning "for my own protection".
    • CommentAuthorLakes
    • CommentTimeJul 24th 2015
     
    Military + Deep Learning = Terminator - SooN.
    •  
      CommentAuthorAngus
    • CommentTimeJul 24th 2015
     
    Can't be SooN enough..
    •  
      CommentAuthorTrim
    • CommentTimeJul 24th 2015
     
    I don't remember writing Urban's two part article on waitbutwhy.

    My memory must really be going.

    If I wrote it it must have been really good.
  2.  
    Deep Learning: Intelligence from Big Data
    https://www.youtube.com/watch?v=czLI3oLDe8M

    Convergence: brain, wavelets, deep learning

    Hmm. Do I go buy an nVidia Tesla Z GPU and start trading again?
  3.  
    •  
      CommentAuthorTrim
    • CommentTimeJul 25th 2015
     
  4.  
    The hypothesis is that GABA needs boosting. Surely that could be tried on autistic humans?
    •  
      CommentAuthorTrim
    • CommentTimeJul 25th 2015
     
    It would be interesting to know why some autistic people are so brilliant at certain things, could all of them be so enhanced?
  5.  
    Distilling the Knowledge in a Neural Network
    http://arxiv.org/abs/1503.02531

    The idea is to train on a massive ensemble of nets, but to use a much smaller network for the actual work of categorisation and generalisation. I was wondering how one would best go about "snipping out a neuron" from a serial chain. In deep learning, there can be up to 20 layers, so such a reduction might be a decent idea.

    I suspect this is a crazy idea, because much of the power of NNs arises from the nonlinearity of the transfer function of each neuron. It seems that these days a simple rectifier with bias is used; i.e. for input x, output y, weight w and bias b, the nonlinear transfer function is y = max(0, w*x - b). Concatenating two of these we get y = max(0, w*y' - b), where y' = max(0, w'*x - b'). In order to "snip out" the middle neuron, we need y = f(x).
  6.  
    Deep Learning and Robotic Control
    https://www.youtube.com/watch?v=EtMyH_--vnU

    What you'd want from this is to train on some humungous net that couldn't possibly be associated in a practical fashion with a real robot - and needn't be. Distillation (see above), or some variant thereof, takes the essence of the net's knowledge and stuffs it into the robot's little computer. No need for a cloud.
  7.  
    http://waitbutwhy.com/2015/01/artificial-intelligence-revolution-2.html
    If you're the first to build an ASI, be sure to build at least two at the same time
    •  
      CommentAuthorTrim
    • CommentTimeJul 26th 2015
     
    I certainly didn't write that, unfortunately.
    •  
      CommentAuthorTrim
    • CommentTimeJul 26th 2015
     
    To avoid extinction humans will have to merge with machine or should I say assimilate with?
    •  
      CommentAuthorAngus
    • CommentTimeJul 26th 2015 edited
     
    Posted By: Andrew Palfreyman


    'Scuse me. Sorry.
    Just so I can read it on this laptop.
    •  
      CommentAuthorTrim
    • CommentTimeJul 26th 2015
     
    NBF

    If Drew can last another twenty years.

    Update on the leading edge of brain implant research and Kurzweils predictions for brain implants in twenty years.

    http://nextbigfuture.com/2015/06/kurzweil-brain-implants-prediction-for.html
  8.  
    Is there room? I'd think the cavity calcified by now.
    •  
      CommentAuthorTrim
    • CommentTimeJul 26th 2015
     
    Remove the tobacco soot and there should be enough room.
  9.  
    Posted By: TrimRemove the tobacco soot and there should be enough room.

    By Jove, I think you're right!

    •  
      CommentAuthorTrim
    • CommentTimeJul 26th 2015
     
    That pic even looks like Drew.