Mb instead of having a fixed annealing schedule, take random step (or jump/reset) in inverse proportion to how good the current optimum is in units of estimated std devs above the estimated mean
--
![[Home]](http://www.bayleshanks.com/cartoonbayle.png)
