Retraining as Approximate Bayesian Inference

TL;DR

This paper reinterprets model retraining as an approximate Bayesian inference process, proposing a decision-theoretic framework that optimizes retraining policies based on a cost minimization approach, leading to evidence-based, auditable triggers.

Contribution

It introduces a novel decision-theoretic framework for retraining policies, framing retraining as approximate Bayesian inference under computational constraints.

Findings

Provides a formal decision-theoretic model for retraining

Defines 'learning debt' as the gap between belief and deployed model

Offers evidence-based, auditable retraining triggers

Abstract

Model retraining is usually treated as an ongoing maintenance task. But as Harrison Katz now argues, retraining can be better understood as approximate Bayesian inference under computational constraints. The gap between a continuously updated belief state and your frozen deployed model is "learning debt," and the retraining decision is a cost minimization problem with a threshold that falls out of your loss function. In this article Katz provides a decision-theoretic framework for retraining policies. The result is evidence-based triggers that replace calendar schedules and make governance auditable. For readers less familiar with the Bayesian and decision-theoretic language, key terms are defined in a glossary at the end of the article.