1. 2
  1.  

  2. 2

    re video. Do you have slides or a transcript for this?

    re first line about limited domain/application. I barely know anything about this subject. The Humies do make me wonder if you’re right about how limited it is, though. I figured, like deep learning, it’s more limited by your hardware resources and time available than anything else.

    1. 2

      Slides here → https://speakerdeck.com/rakhim/genetic-programming-is-waiting-for-better-tools

      I barely know anything, too, this talk was a lot of assumptions and just thoughts. I wonder what you mean by “limit” in this context. Deep learning being limited by computing resources or time means that with enough resources it can do what?

      1. 1

        Thanks for the slides. I’ll check them out later. The comment about limited was in the Youtube description:

        “Genetic programming seems to be applicable in a very limited domain.”

        The one thing I remember most about GP was the Humies. It seemed like they applied it to a lot of areas. So, I was asking what made you think the highlighted statement was true.

        1. 1

          Just finished the slides. Ok, yeah, GP isn’t doing whole applications at the moment. That’s program synthesis. It’s definitely extremely limited in practice compared to what programmers do. Personally, I think it will be done iteratively like humans do it even if evolved. Gotta somehow do multiple layers of decomposition. Even the deep learning systems seem to do that inside.

          Favorite part is the Ferris Wheel of Software Engineering. That was great. :)

      2. 2

        They missed the chance to title this “Genetic programming needs better tools to evolve around it”

        1. 1

          Tools don’t evolve blindly through natural selection, they’re carefully bred!

          1. 3

            You suggest that they are intelligently designed?

            1. 2

              Jury is still out when it comes to a lot of software…