Universal Adversarial Attacks on Neural Networks for Power Allocation in\n a Massive MIMO System

2021/10/10 by Pablo Millán Santos, B. R. Manoj, Santos, Pablo Millán +5 · 1 citation
Computer Science · Biochemistry, Genetics and Molecular Biology · #Wireless Signal Modulation Classification #Adversarial Robustness in Machine Learning #Bacillus and Francisella bacterial research

paper · pdf · doi:10.48550/arxiv.2110.04731

Abstract

Deep learning (DL) architectures have been successfully used in many\napplications including wireless systems. However, they have been shown to be\nsusceptible to adversarial attacks. We analyze DL-based models for a regression\nproblem in the context of downlink power allocation in massive\nmultiple-input-multiple-output systems and propose universal adversarial\nperturbation (UAP)-crafting methods as white-box and black-box attacks. We\nbenchmark the UAP performance of white-box and black-box attacks for the\nconsidered application and show that the adversarial success rate can achieve\nup to 60% and 40%, respectively. The proposed UAP-based attacks make a more\npractical and realistic approach as compared to classical white-box attacks.\n

Cited by

• FGGM: Formal Grey-box Gradient Method for Attacking DRL-based MU-MIMO Scheduler

Related

• Adversarial Attacks on Deep Learning Based Power Allocation in a Massive MIMO Network
• Downlink Power Allocation in Massive MIMO via Deep Learning: Adversarial Attacks and Training
• Adversarial Attacks against Deep Learning Based Power Control in Wireless Communications
• Defending Adversarial Attacks on Deep Learning Based Power Allocation in Massive MIMO Using Denoising Autoencoders
• Practical Adversarial Attacks Against AI-Driven Power Allocation in a Distributed MIMO Network