Unverified50% confidenceOpinionExact time
Previous decoding methods (such as top-k, top-p, and temperature sampling) are mostly based on simplified assumptions about the logit distribution, but these assumptions are rarely verified.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedEvaluating perception systems using only coarse-grained binning by weather severity without considering object-level data leads to misleading conclusions about detector robustness.62% similarUnverified概率校准的常见方法包括温度缩放（temperature scaling）、Platt scaling和等渗回归（isotonic regression），需要在一批带标注的真实数据上拟合校准参数。62% similarUnverifiedPure quantile regression does not provide finite-sample coverage guarantees, and a conformal calibration step must be combined to obtain rigorous statistical meaning.60% similarUnverifiedDuring evaluation, per-case experiment diffs should be examined rather than only looking at changes in aggregate scores, in order to pinpoint regressions to specific cases58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1023607API
curl https://kongchang.com/api/v1/knowledge/claims/1023607MCP
get_claim(id=1023607)