I'm looking for references for #DeepLearning, #ContinualDeepLearning, on very long timescales, specifically #LossOfPlasticity as in Rich/Shibhansh's talk here: https://www.youtube.com/watch?v=p_zknyfV9fY Anyone have any?
I'm in your talk, live-tooting it. Don't let the activity fool you, I'm listening.
Reinforcement Learning researcher, general enthusiast. Finishing PhD after exploring many alternate time streams.
I'm looking for references for #DeepLearning, #ContinualDeepLearning, on very long timescales, specifically #LossOfPlasticity as in Rich/Shibhansh's talk here: https://www.youtube.com/watch?v=p_zknyfV9fY Anyone have any?
Hey fellow dual-account person. But you've explicitly noted you're not using the one here.
But I will say it's much handier finding fellow researchers when you're sitting on the same instance as them. #MastodonSetup
Getting distracted with mastodon setup, so I went and made a dual account. Here for all-science, @annakoop@octodon.social for science and general mayhem.
Will I regret this? PROBABLY! Will I experiment anyway? ALWAYS.
DAGs are too easy, cycles are where it's at.
@glialfire I'm trying the dual-account thing myself for now. We'll see how it goes!
@cmaclell I think that's an important point. I like framing it as #transfer learning, because you have a network that has settled pretty solidly into a particular basin of the parameter space, based on the data it was trained with. This is VERY different than a system that has abstracted generalizable, relevant concepts, which is what I think people want to be true in #zeroshot learning.
@thegradient Maaaaybe but then how do you decide which ones? Vote on it? How narrow is the scope?
Some will happen naturally with following.
@Techronic9876 If I understand correctly, the federated feed is still limited to the union of my instance's follows (see here: https://sigmoid.social/@thegradient/109354173438604962)
Maybe there could be an instance with an account that just follows all known accounts? Then your account on that instance would have twitter-equivalent federated timeline.
Ugh, let's not do that on @sigmoid.social though.