Why these three criteria: negative log evidence equals the area under the loss curve

BlackHC · x · 2026-09-09

The theoretical justification: for an ideal Bayesian learner, using the chain rule of probability, negative log evidence is exactly the area under its loss curve. Hence the three criteria — validation loss estimates the endpoint, CLML is a tail area, and (Bayesian) evidence scores the whole trajectory.

Related event: Bayesian Model Selection Explains Why Proxy-Scale Winners Fail at Target Scale(10 posts)→

Original post →

More from Research

Research channel →