Where does boosting collide?

In machine learning, boosting is an ensemble method that builds a strong predictor out of many weak learners, each only slightly better than chance, trained in sequence so that later learners concentrate on the cases earlier ones got wrong. Schapire's 1990 The strength of weak learnability proved that a weak learner can in principle be boosted into an arbitrarily accurate one, the result that launched the family later realised as AdaBoost and gradient boosting. The referent is an algorithmic procedure for reweighting data and aggregating classifiers, unrelated to the behavioural-policy idea of building competence beyond the shared word.[1]

Unit

The combined prediction of a weighted ensemble of weak learners

Recovery means

Not applicable

Measured by

Tracked by the ensemble's training and test error as successive weak learners are added

In behavioural public policy, boosting is an intervention approach that aims to expand people's own competences (their skills, knowledge, and decision tools) so they can choose better for themselves, as opposed to nudging, which rearranges the choice environment while leaving competences untouched. Hertwig and Grüne-Yanoff's 2017 Nudging and boosting: Steering or empowering good decisions drew the distinction and named boosting as the empowerment alternative. The referent is a class of policy interventions defined by what they build in the decision-maker, a different construct from an ensemble-learning algorithm despite the shared term.[2]

Unit

The lasting competence a person gains to decide well on their own

Recovery means

Not applicable

Measured by

Evaluated by whether a taught competence persists and transfers beyond the intervention