Unverified50% confidenceFactExact time
IQR is insensitive to extreme values and is particularly suitable for describing target area distributions that may exhibit a long tail or even zero values after cropping; when augmentation causes the targets in some samples to nearly disappear, the standard deviation is significantly inflated by outliers, whereas IQR still robustly reflects the dispersion of the main body of samples.
1
Sources
50%
Confidence
Long-term
Relevance
11/1/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedWhen 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.73% similarUnverifiedCompared to the Z-score method based on standard deviation, the IQR method is not affected by extreme values and is better suited for handling skewed distributions.72% similarUnverified当综合评估指数中顶尖模型得分接近天花板时会出现基准饱和问题，60分对59分的差距在统计学上通常处于误差范围之内70% similarUnverifiedThe endpoint metric used in the evaluation is the difference of 'model minus uniform control group', where a positive value represents performing worse than the control.70% similarUnverifiedCertain AI capabilities are nearly zero when model scale is below a specific threshold, then rise sharply once the threshold is crossed, exhibiting a nonlinear 'phase transition' characteristic.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1123446API
curl https://kongchang.com/api/v1/knowledge/claims/1123446MCP
get_claim(id=1123446)