Unverified50% confidenceOpinionExact time
When the within-group reward distribution is severely skewed, the mean is pulled off by extreme values, causing distorted advantage estimates; in such cases, the median or a trimmed mean may be more robust.
1
Sources
50%
Confidence
Long-term
Relevance
10/28/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedWhen the confidence distribution is highly imbalanced, proportional allocation significantly reduces estimation error compared with equal allocation per stratum.74% similarUnverifiedWhen the success rate is extremely low, almost all sample rewards are identical, resulting in extremely high policy gradient variance and very low signal-to-noise ratio, preventing the model from obtaining effective update directions.74% similarUnverifiedFor categorical policies, as the output distribution becomes increasingly sharp, the expected score-gradient norm approaches zero, resulting in the weakest updates for tokens that are confident but incorrect.74% similarUnverifiedPercentage tolerance makes inputs of different magnitudes comparable and decouples thresholds from numerical scale, but the cost is that small absolute fluctuations in small numbers are amplified into high percentages, while large absolute changes in large numbers may only produce moderate percentages.72% similarUnverified测试时计算扩展的边际收益并非线性，存在明显的收益递减区间72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1109170API
curl https://kongchang.com/api/v1/knowledge/claims/1109170MCP
get_claim(id=1109170)