Jul 2026
Best-Arm Identification with Generative Proxy
It is proved that PROBE is $\delta$-PAC and attains the known-correlation oracle sample complexity up to a constant multiplicative factor and a constant additive calibration cost and the guarantee extends to the $(\epsilon,\delta)$-PAC setting under minimal changes to the algorithm.
Tianyi Ma, Han-Zhang Qin, Ruihao Zhu et al.
· arXiv.org · 1 citation