And if you let me tinker a bit with your model of how inheritance works (and biology is quite unclear about that), exponential convergence may be possible.
-
-
Odgovor korisnicima @KordingLab @blake_camp_1 i sljedećem broju korisnika:
Right (gain's wrt population size although there's a bound that places Statistical query learning unless you want Lamarkism https://www.sciencedirect.com/science/article/pii/S0022000012000232 …) So my essential point is, if we look at integration methods which _explicitly_ use gradients, what are nuances, utility?
1 reply 0 proslijeđenih tweetova 1 korisnik označava da mu se sviđa -
Odgovor korisnicima @sir_deenicus @KordingLab i sljedećem broju korisnika:
For utility it's to be able to safely take larger step sizes. So now there's a link in the analogy to allow reasoning back and forth. I'm currently skeptical about how large these stepsizes can be. More interested in how evolution metalearns operators such that space is smoother!
1 reply 0 proslijeđenih tweetova 2 korisnika označavaju da im se sviđa -
Odgovor korisnicima @sir_deenicus @blake_camp_1 i sljedećem broju korisnika:
I am firmly in lamarckian territory.
1 reply 0 proslijeđenih tweetova 3 korisnika označavaju da im se sviđa -
Odgovor korisnicima @KordingLab @blake_camp_1 i sljedećem broju korisnika:
You think it's possible for animals to store knowledge in their genome and transmit what they learned to the next generation?
0 proslijeđenih tweetova 1 korisnik označava da mu se sviđa -
Odgovor korisnicima @sir_deenicus @KordingLab i sljedećem broju korisnika:
Would u disagree that it's possible to pass on a measure of the usefulness of innate priors w.r.t intra-lifetime curriculum?
1 reply 0 proslijeđenih tweetova 2 korisnika označavaju da im se sviđa -
Odgovor korisnicima @blake_camp_1 @KordingLab i sljedećem broju korisnika:
Only as far as Baldwin effect captures. Which interestingly enough, disfavors learning unless the cost (such as from instability or search space complexity) make it worthwhile.
1 reply 0 proslijeđenih tweetova 2 korisnika označavaju da im se sviđa -
Odgovor korisnicima @sir_deenicus @blake_camp_1 i sljedećem broju korisnika:
No, the Baldwin effect promotes learning! With the Baldwin effect, animals that learn evolve faster than those that don't, so they have a real advantage in the long run.
1 reply 0 proslijeđenih tweetova 3 korisnika označavaju da im se sviđa -
Odgovor korisnicima @tyrell_turing @blake_camp_1 i sljedećem broju korisnika:
There is a bit more nuance than that, which surprised me when I first encountered it. There's some disagreement but general consensus here is, learning is good at first but the evo trajectory is generally to reduce the amount of required learning. Related: canalizationpic.twitter.com/LiuN7OtyHV
1 reply 0 proslijeđenih tweetova 2 korisnika označavaju da im se sviđa -
Odgovor korisnicima @sir_deenicus @blake_camp_1 i sljedećem broju korisnika:
I think you misinterpreted that result... The Baldwin effect does indeed promote ever faster learning to the point that it is so fast and requires so little input it's essentially hardwired. So, in that sense, it promotes things to be nigh hardwired if they can be learned.
0 proslijeđenih tweetova 4 korisnika označavaju da im se sviđa
But that doesn't mean learning is selected against! Quite the opposite. It means learning species will usually out compete non-learning species, and they will develop more hardwired behavior at a faster rate. So, learning will be strongly selected for.
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.