[KongchangAI]
Conceptbinned calibration / 长度条件校准

Bin-Wise Calibration

A calibration method that bins texts by length and calibrates the detection threshold separately within each length interval using conformal p-values, so that false positive rate control is conditional on length.

Timeline (last 90 days)

Nov 10

The research team proposed a binning calibration scheme: divide texts into intervals by length, and within each interval calibrate the threshold separately using conformal p-values.

Unverified50%
Nov 10

On three zero-shot detectors, three corpora and two open-weight generators, averaged over 300 random document-level repartitions, binning calibration stabilized the average false positive rate at 5.0% in every detector-length combination from 10 tokens to 200 tokens.

Unverified50%

All Facts (2)

Source Articles