Re:Akaike information criterion
by bendysnowflakes on Monday January 08, 2018 @09:36PM
(#16103060)
Yeah, I don't mean to make it sound like there's no good reason for CV... I came across as sort of harsh. I just feel like CV got taken up in application a lot because it makes intuitive sense but with relatively little in terms of theoretical scrutiny, especially relative to other alternatives. When it has been compared to alternatives, the comparisons are sort of weak in that they often involve strawman or unrepresentative assumptions. CV involves certain assumptions itself, for example, in terms of choices, that affect results in sort of unpredictable ways sometimes. I'm a big fan of bootstrapping too and have often felt that that avenue hasn't been explored as much as it could be.
AIC is controlled by the number of parameters and the quality of fit, but it is known in theory, as well as in practice that for many models, for example, neural networks, linear regression with additive iid Gaussian noise, the norm of the parameters matter more for prediction accuracy than the sheer number of parameters.
Not claiming that CV is the best model selection criteria, just listing out a few pluses for CV. The last word on model selection is yet to be spoken. It goes right to one of the core questions about model fitting and prediction.