similar flatness-seeking implicit biases have been discovered in other stochastic optimization algorithms; here we reveal their different underlying mechanisms despite similar outcome, demonstrating the unique properties of evolutionary dynamics.
The paper’s behind a paywall, but from the abstract it seems like the main claim is that they found another mechanism for an already well-established result (essentially, that drunkards develop an aversion to strolling near cliffs). It doesn’t really suggest how their findings differ from earlier studies of quasispecies models, though.
Basically, what the paper looks to be saying is that once a species reaches that optimal survival peak and traditional natural selection stops pushing it up the hill, the shape of that peak itself becomes the selection pressure. When different genetic combinations can produce the same winning traits, the system naturally slides away from the edges where a single mutation might ruin everything and drifts toward the center where the genetic code is most robust and forgiving.
Yeah, that sounds like the usual mechanism (going back at least to this paper from 25 years ago, which coined the term “survival of the flattest”).
I just got the impression for the abstract that they might have found something significantly different.
Ah yeah, don’t think there’s anything really new here.
