Buffl

VL0WDHLR

df
von diana F.

Warum ist „einfacher“ besser? Was ist die Begründung für Ockham‘s Razor?

• There‘s no logical justification for Ockham‘s razor (i.e., there is no proof that, even ceteris paribus, the simpler theory is true/closer to truth).

• E.g. evolution depends on random mutations, crystallized randomness: no reason to assume that simpler is (always) closer to truth.

• Can be seen as a probabilistic argument: Of two explanations with the same explanatory power, the simpler has a higher probability of being true / closer to the truth.

• Can be seen as a pragmatic argument: When the predictive success is the same, it is easier & more tractable for us to retain the simpler theory, without any practical loss.

Saves cognitive capacity, less teaching, less memorizing. Do not waste time on irrelevant stuff.

• See it more as a heuristic, not a law

• Concerning in-sample fit, the more complex model always outperforms the simpler model (or is in the boundary case at least equally good). But: when we switch to out-of-sample predictions, the simpler model can be better (even if you do not believe in „truenulls“)

• Does ‚better‘ mean ‚more true‘ or ‚more useful‘? If the latter, Ockham‘s razor can be justified.

• As an antidote to Ockham: Epicurs’ Principle of Multiple Explanations states: “If several theories are consistent with the observed data, retain them all.” (http://cage.ugent.be/~ci/Epicurus.html)

• David MacKay’s online book ITILA (http://www.inference.phy.cam.ac.uk/itila) chapter 28

(http://www.inference.phy.cam.ac.uk/mackay/itprnn/ps/343.355.pdf) gives the clearest justification fo

Author

diana F.

Informationen

Zuletzt geändert