Delving into Cryptanalytic Extraction of PReLU Neural Networks

Yi Chen ID, Xiaoyang Dong ID, Ruijie Ma ID, Yantian Shen ID, Anyu Wang ID, Hongbo Yu ID, Xiaoyun Wang (🖂) ID · 2025

Read Delving into Cryptanalytic Extraction of PReLU Neural Networks by Yi Chen ID, Xiaoyang Dong ID, Ruijie Ma ID, Yantian Shen ID, Anyu Wang ID, Hongbo Yu ID, Xiaoyun Wang (🖂) ID (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-2509-16620.txt.

Description

Yi Chen and co-authors attack a type of neural network to steal its settings, in a 2025 security paper. Stealing a model's inner values has been studied since the early days of the field. Most work targeted networks with a simple bend in their activation. This team target networks that use a leaky variant with a slope below zero.

Their attack works from raw model outputs. They extend it to cases where only the top few scores are shown. Tests on varied networks, some trained on handwritten digits, confirm the attack works end to end. The work is claimed as the first real theft of this network type. Model owners should note the risk.

Catalog record

CREATORYi Chen ID, Xiaoyang Dong ID, Ruijie Ma ID, Yantian Shen ID, Anyu Wang ID, Hongbo Yu ID, Xiaoyun Wang (🖂) ID
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE83 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-16620

Browse the full library · About the Tower of Babel