[KongchangAI]
ConceptCRPS / 连续排名概率评分

Probabilistic Scoring

A metric measuring how well a predictive distribution matches observed actual values. It penalizes mean bias and insufficient ensemble spread, and is used to assess the reliability of probabilistic forecasts.

Timeline (last 90 days)

Nov 10

On seventeen synthetic targets, guidance more than halves the median bias of the 1st and 99th percentiles of daily temperature, improving probability scores and reliability.

Unverified50%
Oct 19

In resource-constrained settings, smaller models can be used to improve NLI performance through probability scoring rather than direct classification.

Unverified50%

All Facts (2)

Source Articles