3 ms·
In evolutionary algorithms, fitness climbs rapidly from a random start, then tends to level off over generations. The "split-and-hit-the gas" dynamic makes sens
by vannevar 1y ago
In evolutionary algorithms, fitness climbs rapidly from a random start, then tends to level off over generations. The "split-and-hit-the gas" dynamic makes sense if you think of a given phenotype occupying a certain phase space. If a sudden divergent mutation throws it into a new phase space, it's sort of resetting the evolutionary clock: fitness in the new dimensions of the space increases more rapidly, amplifying the divergence from the other evolutionary branch.
- ACCount37 1y agoNature has one thing most evolutionary algorithms don't: environmental changes. That has a way of stirring up the pot. Which is not at all to say that you're wrong. A lot of the known major environmental changes were brought upon by other creatures - sometimes creatures that found some weird new way of doing things, and upended the environment with it.
- vannevar 1y agoYou are 100% right. In nature, the fitness landscape is not fixed, it is dependent on the co-evolution of other organisms at every scale. So the computational model is too simple to be a reliable guide. But in this case, if you assume constant fitness over some delta-t (kind of like calculus), I think it's still a plausible explanation for the divergence phenomenon in particular. Just a hypothesis, though, for sure.