Performative Risk Control: Calibrating Models for Reliable Deployment under Performativity

Victor Li, Baiting Chen, Yuzhen Mao, Qi Lei, Zhun Deng · 2025

Read Performative Risk Control: Calibrating Models for Reliable Deployment under Performativity by Victor Li, Baiting Chen, Yuzhen Mao, Qi Lei, Zhun Deng (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-24097.txt.

Description

Victor Li and co-authors tackle a calibration problem that arises when predictions change the world they predict in this 2025 paper. A loan model, for example, can alter who repays, so risk targets set on old data may fail once the model is deployed. Standard calibration assumes the data stays fixed, which performative settings break.

Their framework, Performative Risk Control, refines calibration step by step while keeping risk guarantees at every stage. They study several risk measures and tail bounds to support the math. Tests on credit default prediction show the method holding its guarantees in practice. The authors present it as a guard against strategic gaming of deployed models.

Catalog record

CREATORVictor Li, Baiting Chen, Yuzhen Mao, Qi Lei, Zhun Deng
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE74 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-24097

Browse the full library · About the Tower of Babel