For fixed-confidence multiple change point identification under bandit feedback, the paper derives instance-dependent lower bounds and an asymptotically optimal Track-and-Stop variant (MCPI) that samples near each jump in proportion to one over the jump size squared.
Hence, from (57), if Nj 4 (∆(j) − 2r(t))2 (58) holds then we will never play actions x∗ (j) or x∗ (j) + 1 when tracking again
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ML 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Fixed-Confidence Multiple Change Point Identification under Bandit Feedback
For fixed-confidence multiple change point identification under bandit feedback, the paper derives instance-dependent lower bounds and an asymptotically optimal Track-and-Stop variant (MCPI) that samples near each jump in proportion to one over the jump size squared.