A hybrid online-plus-offline preference optimization algorithm, HPO, provably needs fewer samples than pure online or offline RLHF in linear MDP settings.
For completeness, we first describe the preference model below: BTL-based Pairwise Preference (Dueling) Model:Consider a decision spaceD
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Hybrid Preference Optimization for Alignment: Provably Faster Convergence Rates by Combining Offline Preferences with Online Exploration
A hybrid online-plus-offline preference optimization algorithm, HPO, provably needs fewer samples than pure online or offline RLHF in linear MDP settings.