HSBC learns rewards from preference batches by conservatively cutting the hypothesis space with a voting threshold, tolerating erroneous labels up to a bound set by a parameter gamma.
K., Anil, R., and Koren, T
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
-
Robust Reward Alignment via Hypothesis Space Batch Cutting
HSBC learns rewards from preference batches by conservatively cutting the hypothesis space with a voting threshold, tolerating erroneous labels up to a bound set by a parameter gamma.