3 ms·
I just loved this bit in the paper, could be so easily taken on so many tangents!: ""Delayed Gratification is used to evaluate the ability of each algorithm un
by danwills 2y ago
I just loved this bit in the paper, could be so easily taken on so many tangents!:
""Delayed Gratification is used to evaluate the ability of each algorithm undertake actions that temporarily increase Monotonicity Error in order to achieve gains later on. Delayed Gratification is defined as the improvement in Sortedness made by a temporarily error-increasing action.""
Is it slightly analogous in some ways to the avoidance of getting stuck in local maxima perhaps?
- vindex10 2y agoOr maybe the fact that the path of minimal effort != sequence of paths with locally simplest steps.
- danwills 2y agoI'd love to know more about what you mean! Is it that with a bit more upfront investment (or 'delayed gratification'), a system might be able to find shorter (or less energy-intensive) paths through the space that they are navigating to get to their targets? I think it certainly does seem like when there is essentially some 'computational slack' ('extra line' say, slightly over-provisioned or just set-to-explore more) I'd guess there's a good chance that that could yield a better (cheaper/shorter) result than a brute-force (minimal-effort, perhaps more technically 'efficient') solution? Unfortunately I don't really feel like I really went anywhere with what I said, but your very short comment made me wonder many things about what you meant! Appreciated!