Beyond Leave-One-Out : Private and Robust Contribution Evaluation in Federated Learning DOI: XXXXXXX.XXXXXXX CCS: Computing methodologies Machine learning Distributed algorithms CCS: Security and privacy Privacy-preserving protocols CCS: Computer systems organization Distributed architectures

Delio Jaramillo Velez, Gergely Biczok, Alexandre Graell i Amat, Johan Ostman, Balazs Pejo · 2026

Read Beyond Leave-One-Out : Private and Robust Contribution Evaluation in Federated Learning DOI: XXXXXXX.XXXXXXX CCS: Computing methodologies Machine learning Distributed algorithms CCS: Security and privacy Privacy-preserving protocols CCS: Computer systems organization Distributed architectures by Delio Jaramillo Velez, Gergely Biczok, Alexandre Graell i Amat, Johan Ostman, Balazs Pejo (2026) 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-2602-21721.txt.

Description

Cross-silo federated learning lets organizations train models together without sharing raw data, yet client updates can still leak sensitive facts through inference attacks. Secure aggregation hides individual updates, but that hiding blocks contribution scoring, which matters for fair rewards and for spotting weak or malicious participants. Shapley value scoring clashes with secure aggregation, while leave-one-out scoring is crude and relies on self-evaluation.

Delio Jaramillo Velez and four co-authors introduce two marginal-difference scores that work under secure aggregation. Fair-Private meets standard fairness axioms, while Everybody-Else drops self-evaluation and resists manipulation. Both carry theoretical guarantees for fairness, privacy, robustness, and efficiency, and on medical image data and CIFAR10 they beat baselines, track Shapley rankings better, and improve both downstream performance and misbehavior detection.

Catalog record

CREATORDelio Jaramillo Velez, Gergely Biczok, Alexandre Graell i Amat, Johan Ostman, Balazs Pejo
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE42 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-21721

Browse the full library · About the Tower of Babel