Understanding neural networks


It’s an old idea. It’s an idea that came from neuroscience. Does a though live in an individual neuron? Or do all neurons in the human brain participate in all ideas? And it’s been very hard to test in humans because you can’t put a probe on every single neuron in the human brain. In an artificial neural network, we have this luxury of being able to look at everything that’s going on. [music] Our project is really asking the question: “What is a neural network learning inside?” And we study a specific kind of network called a GAN. That’s a generative adversarial network. We tell it, “Imagine and image that you haven’t seen that looks like these million other images”. [music] The surprising result of the project is that neural networks actually show evidence of composition. And so the question is, “How the heck is it doing it?” [music] If it’s just memorizing, then it’s approaching things the way we normally program computers to do things, right? If it’s composing, it’s sort of a sign that it’s thinking in a more human-like way, that it’s understanding the structure of the world. [music] But correlation is not the same as causation. It could be that neuron that correlates with trees is actually what the neural network is using to think about the color green. So how do we know the difference? And just like those individual neurons that correspond to trees or doors, We found that there are individual neurons that actually correlate with these visible bugs, with these visible artifacts. So that was really surprising to us. Because not only is the network sifting through things, and sorting out things that make sense, It’s also sifting and assigning the things that don’t make sense to their own variables as well. And so it was really surprising to us that we go into a neural network and do a certain type of brain damage, right? Basically perform a lobotomy on these twenty neurons, and instead of doing damage to the network, we actually got the network to perform better. And so why is it that a network actually has neurons in it that cause problems? Are mistakes an important part of learning? It’s one of the mysteries that we uncovered. We don’t know the answer to that. But I think that there’s more profound reasons to be interested in this beyond the ancient puzzle of, “How does thinking work?” and, “How do humans work?” Because we’re also using these AI’s to build our future world, to build our future societies and it’s important that we are able to understand, anticipate, and control the world that we create, and as long as we don’t really understand what rules they’re applying inside, we’re not going to be able to do that. And so, I think, I don’t know, I think it’s the most important thing in the world to study this kind of thing. [music]

, ,

Post navigation

51 thoughts on “Understanding neural networks

  1. Noting the difference between causation and correlation is good although a consideration not applied anywhere nearly enough. Also, how humans work in by no means a mystery.

  2. Do the so-called lobotomy improve the network in all cases? Does it only work in some instances? How is this different from a Dropout layer in the neural network?

  3. "Imagine an image that you haven't seen that looks like one of these images."

    And, success!!

    Yep. I, for one …

    (Really, I do. Bring on the Singularity, soft-spoken genius. Go, go, go!,!!,!!!!)

    Moreover, the "imagination" wasn't the point. Yep. ENABLE THE SINGULARITY!

  4. maybe it's cuz when neuron performs a function and then those 20 neurons are erased, it has more space to perform the various function better. Like how workers can't work efficiently in cluster areas. maybe some breathing room is required and then other workers just take on the job, making the work to be even cheaper to produce while performing it (in this case, faster). But cut away too much and you just have massive cell death. So the second question will be is what's the sweet spot?

  5. Their smiling is fake..it's something very mysterious..they wanna create an environment for our society and control that directly with our brain..I smell something else than what's being shown

  6. 2:56 Understand AI logic first. That is really important. If we still blindly 'using' AI, not only the helm is not ours, the ship of human civilization will be draged into AI's powful wirl. THE PROBLEM NOW of the world is, not too many AI growners now in tech companies are careful and rational like him.

  7. I was think about something the other day, "like coding humans" type stuff. What can to mind type stuff, was when numbers equal colors and shades of text. Like windows paint basically, but using numbers or digits to remeber or code colors. like what number in 15,000's hindred would make blue, can you see in your head.

    if you think if innosonifigant i want to refer you to the Lisa woo, chart.. it use it coloros and graphics, as medical reference for pain, emotions and feelings… but guess what made that chart? color coding, by number and graphic design in the past ( maybe crayons too so to speak)

  8. sometime, money does grow on trees — decision trees, but if you promote too many things, you'll scare the public away hahaha

  9. Lately I have been so interested in machine learning and I have reading and watching videos and trying to do some things in python. I am truly inspired to keep learning after watching this video. Thank you very much.

  10. New series on Deep Learning for those who want to go deeper: https://www.youtube.com/watch?v=YulgDAaHBKw&list=PLbg3ZX2pWlgKV8K6bFJr5dhM7oOClExUJ

  11. Neural networks are highly likely a core technology of the future. That work is massively important for sure. It's just that nowadays the field is so crowded that smart people should maybe think about more neglected areas.

  12. There are four different theories about how the universe was born before the Big Bang
    The problem is that we need only one theory
    We need one theory that explains the emergence of the universe
    The first theory is the collision theory of cosmic membranes

    This theory says that the birth of this universe was caused by the explosion caused by the collision of membranes every trillion years

    The second theory is that the universe is the hologram world
    It is that the universe is a stereoscopic image of black hole information
    This means that our lives and our self-consciousness are just an illusion
    We are just an illusion and not a reality
    ‏The third theory says that the universe is just a bubble between an infinite number of bubbles of random inflation
    ‏The fourth theory says that the Great Bang created our world 14 billion years ago of white hole energy
    ‏‏But is there scientific evidence of the birth of the universe from this white hole?
    ‏ ‏The Continuing Problem There are four different theories about the birth of the universe
    ‏ ‏You must create a new satellite
    ‏‏In order to discover the oldest light in the universe
    ‏In order to discover the signs (Waves of initial gravity)
    ‏We need to create a number of new satellites dedicated to exploring scientific evidence
    ‏ ‏To confirm the hologram theory or evidence of the collision theory of those cosmic membranes
    ‏Or evidence supporting the hypothesis of inflationary cosmic bubbles

  13. I wish there were more white women in STEM. This video translates that neural networks are just for men or asians.

Leave a Reply

Your email address will not be published. Required fields are marked *