Protecting Floating-Point Computation for DNN Binaries with MBA Obfuscation

2026/07/20 by Yikun Hu, Zichen Zhao, Peixiang Qin +4
#cs.CR

paper · pdf

Abstract

This submission was made prematurely and has been withdrawn by the authors for substantial revision before further dissemination.

Related

• Accuracy Booster: Enabling 4-bit Fixed-point Arithmetic for DNN Training
• NNSplitter: An Active Defense Solution for DNN Model via Automated Weight Obfuscation
• Training DNNs with Hybrid Block Floating Point
• ObfusBFA: A Holistic Approach to Safeguarding DNNs from Different Types of Bit-Flip Attacks
• Poster: E-Graphs and Equality Saturation for Term-Rewriting in MBA Deobfuscation: An Empirical Study