In sequential testing, skipped tests hide decision rewards, forcing any algorithm to incur Ω(T^(2/3)) regret; Explore-Then-Commit matches this, while an entropy-sampling variant with rewards independent of missing data achieves √T regret.
Journal of the ACM (JACM) 67(6):1--42
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Online Learning of Optimal Sequential Testing Policies
In sequential testing, skipped tests hide decision rewards, forcing any algorithm to incur Ω(T^(2/3)) regret; Explore-Then-Commit matches this, while an entropy-sampling variant with rewards independent of missing data achieves √T regret.