ReCIRC: Rectified Conformal Risk Control
ReCIRC reparameterizes conformal risk control so thresholds target conditional risk while keeping finite-sample marginal guarantees.
ReCIRC inverts each input's estimated local risk curve and reparameterizes the calibrated threshold as a shared risk budget before applying standard conformal risk control. It retains CRC's finite-sample marginal guarantee even when the estimated curves are inaccurate, while accurate curves support approximate and, under further conditions, asymptotically exact conditional risk control. Across three synthetic and five real settings, it recorded the lowest average worst-group risk and mean positive group excess while staying near the target marginal risk.
- Preserves conformal risk control's finite-sample marginal guarantee.
- Lowest worst-group risk in all eight reported settings.
- Covers segmentation, classification, and regression tasks.
Full article188 words · extracted from arxiv.org · click to collapse
Many applications of black-box predictive models require controlling task-relevant error rates, such as missed lesion pixels in segmentation or missed labels in multilabel classification. Conformal risk control (CRC; Angelopoulos et al., arXiv:2208.02814) gives distribution-free guarantees for such losses, but it calibrates a single threshold shared by all inputs. Because conditional risk varies with the input, this marginal guarantee often overprotects easy cases and underprotects hard ones. We propose ReCIRC (Rectified Conformal Risk Control), which inverts each input's estimated local risk curve to reparameterize the calibrated threshold as a risk budget $a$ representing a common target conditional risk, and then applies CRC unchanged to the resulting family. ReCIRC retains CRC's finite-sample marginal guarantee regardless of the accuracy of the estimated curves, while accurate curves yield approximate conditional risk control and, under additional conditions, asymptotically exact conditional risk control; they also support a risk-calibration diagnostic. Across three synthetic and five real-data settings spanning segmentation, multilabel and multiclass classification, and regression, ReCIRC attained the lowest average worst-group risk and mean positive group excess in every setting, while maintaining marginal risk close to the target, whereas changes in prediction size were application-dependent.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2609.38112