9 Matching Annotations
  1. Oct 2022
    1. . Deep learning methods rely even less on human knowledge, and use even more computation, together with learning on huge training sets, to produce dramatically better speech recognition systems.

      deep learning thus is a development in the process of the diminishing of human knowledge in emphasis in developing AI

  2. Aug 2019
    1. Gee (2007: 172) describes deep learning as “learning that can lead to real understanding, the ability to apply one’s knowledge, and even to transform that knowledge for innovation.” He argues that pursuing deep learning requires moving beyond learning about – “what the facts are, where they came from, and who believes them” – to learning to be – which involves “design” in the sense of understanding how and when and why knowledge of various kinds is useful for and sufficient for achieving particular purposes and goals. According to Gee (2007: 172)

      This is a great paragraph about deep learning. Not just learning about but "learning to be" which involves design.

  3. Mar 2019
    1. Same-different problems strain convolutional neural networks

      Wow, this is fascinating. I wonder if SD problems could be the next major roadblock for AGI...

  4. Feb 2019
    1. Weight symmetry between the forward and backward passes and delayed error generation are two ofthe most biologically unrealistic aspects of backpropagation.
  5. May 2015
    1. like this sketch

      I like the graph. I have to follow the roots of this unlearning literature. Feeling a legacy of Piaget in this sense of "crisis" or "discomfort" that is required for deep learning.

      I know I throw things when writing, yet there is also a sense of elation and drive.

      I find the or strange. I wonder if It should be and. Many students put in time without effort and get no where.

      I also think you can reach understanding instantaneously (though I think Sam refers to more designed learning than natural learning). I think about Kristeva and the abject. It isn't so much unlearning but a reversion.

      My dabbling in this makes me wonder if we would need tools in the chart or the activity...wait we would be stuck with Engstrom's triangles again.

    2. know more at precisely the same moment that you understand less.

      Or is this a recognition that you have so much more to learn? Is understanding from this framework nothing more than the motivation from greater knowledge?

    3. while not yet being able to meaningfully connect it to things you already know.

      This puts deep learning in the hands of the individual I am beginning to wonder if understanding is something that belongs to the collective. It is too subjective in the individual.

    4. The inability to connect a new piece of information with the world as we already know it--this is a classic problem of the unlearning that is required for deeper learning

      It could be just not encountering enough variations across multiple case studies.

      I also see many parallels to the idea of what we are calling synthesis here.

    5. knowledge while losing understanding

      I agree with this statement but I do not by into the science of unlearning. You are not "unlearning" when your perceptions shift. It is a movement or trajectory.

      I need to explore this more but the field of research in misconceptions is much stronger in the hard sciences. I am not too comfortable with it, but ill-defined and well-defined domains do behave differently. Oops I just anthropomorphized knowledge. Mistake?

      To me the idea that of starting with the learner is wrong is wrong. Deeper learning does not have to begin from here.