Climbing The Wrong Hill


爬山算法:通过寻找附近的upward step可能使得你只是找到local optimal解。相反如果可以随机地放弃一些附近的upward step, 有可能得到global optimal解。

But the lure of the current hill is strong. There is a natural human tendency to make the next step an upward one. He ends up falling for a common trap highlighted by behavioral economists: people tend to systematically overvalue near-term over long-term rewards. This effect seems to be even stronger in more ambitious people. Their ambition seems to make it hard for them to forgo the nearby upward step. 人通常更看重短期回报,并且越有抱负的人更加如此。结果导致这些人更难于放弃最近的upward step.

People early in their career should learn from computer science: meander some in your walk (especially early on), randomly drop yourself into new parts of the terrain, and when you find the highest hill, don’t waste any more time on the current hill no matter how much better the next step up might appear. 所以尤其在初期,应该尽可能多地接触各种信息,然后偶尔让自己尝试新的领域。如果你发现自己出去山顶的话,那么不要在花费更多的时间在这个山顶上,不管下一步看上去多么好(因为可能你已经进入了局部最优解)

comments powered by Disqus