On the Fragility of Data Attribution When Learning Is Distributed

Xian Gao, Bo Hui, Min-Te Sun, Wei-Shinn Ku · 2026

Read On the Fragility of Data Attribution When Learning Is Distributed by Xian Gao, Bo Hui, Min-Te Sun, Wei-Shinn Ku (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-2605-15520.txt.

Description

Xian Gao, Bo Hui, Min-Te Sun, and Wei-Shinn Ku show that credit for data in shared training can be gamed. One player can puff up its own measured credit while keeping overall skill. Their attack slips in small fake batches tuned by latent search, playing on patchy label spread and tester quirks. It lifts the foe's credit and reorders credit for honest players, with no loss of accuracy and no trip of shape-based guards. Credit itself is thus a fresh attack front.

Catalog record

CREATORXian Gao, Bo Hui, Min-Te Sun, Wei-Shinn Ku
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE58832
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.15520

Browse the full library · About the Tower of Babel