Summary
This paper by Georgy Noarov and Aaron Roth (arXiv:2608.13554) studies online probabilistic forecasting of binary outcomes chosen by an adaptive adversary. The authors introduce the Defensive Booster, a defensive forecasting algorithm that simultaneously achieves two previously incomparable guarantees from existing online boosting techniques: (1) like online gradient boosting, its Brier score competes with the best predictor induced by the span of a weak hypothesis class H on every adaptive sequence; and (2) like online weak-to-strong boosting, whenever the realized transcript satisfies a smooth weak-learning condition, its Brier score and randomized classification error match the rate guarantees of online classification boosting. The approach operationalizes the dual view of boosting: persistently high randomized classification error yields an ex-post hard-core certificate that the weak-learning condition fails. A strongly adaptive variant satisfies both guarantees on every time interval. Notably, the algorithm accesses just one weak-class learner, unlike prior online boosting methods that maintain large ensembles, and experiments on synthetic and real data streams show strong predictive performance with orders-of-magnitude faster runtime.
Paper Overview
Research Area: ML
Authors: Georgy Noarov, Aaron Roth
Published: 2026-08-13
arXiv:
2608.13554Abstract
We study online probabilistic forecasting of binary outcomes chosen by an adaptive adversary. Given an online learning algorithm for a weak hypothesis class H, we would like to efficiently obtain two incomparable guarantees that existing online boosting techniques provide separately. Online gradient boosting competes in Brier score with the best predictor induced by the span of H on every sequence, but promises nothing when the span does not contain an accurate predictor. Online weak-to-strong boosting drives classification error to zero under a weak-learning condition, but promises little when that condition fails. We give a simple defensive forecasting algorithm, the Defensive Booster, that obtains both guarantees. On every adaptive sequence, its Brier score is competitive with the best prediction induced by the span of H at the same rate as online gradient boosting; simultaneously, whenever the realized transcript satisfies the smooth weak-learning condition, its Brier score and randomized classification error satisfy the same rate guarantee as online classification boosting. This is achieved by operationalizing the "dual view" of boosting: When the algorithm's randomized classification error is persistently high, its mistake weights form a smooth reweighting on which every weak hypothesis has low edge, yielding an ex-post hard-core certificate that the weak-learning condition fails. We also develop a strongly adaptive variant, which satisfies both guarantees on every time interval. The Defensive Booster is very efficient: it accesses just one weak-class learner, whereas the prior online boosting methods we compare against maintain large weak-learner ensembles. Experiments on synthetic and real data streams demonstrate its strong predictive performance (sometimes substantially improving over all prior baselines) coupled with orders-of-magnitude faster runtime.
Key Contributions
- Dual guarantees: The Defensive Booster combines the always-on Brier score guarantee of online gradient boosting with the weak-learning-condition rate guarantees of online classification boosting.
- Ex-post certificates: Persistently high randomized classification error produces a smooth reweighting certifying that the weak-learning condition fails.
- Strongly adaptive variant: Both guarantees hold on every time interval.
- Efficiency: Accesses only one weak-class learner instead of maintaining large weak-learner ensembles, achieving orders-of-magnitude faster runtime with strong empirical performance.
---
*Auto-collected on 2026-08-15*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/178633497