Algorithms to Live By: The Computer Science of Human Decisions

by Brian Christian

But when the odds of estimating them all correctly are low, and the weight that the model puts on those untrustworthy quantities is high, then an alarm should be going off in the decision-making process: it’s time to regularize.