Over-The-Air Adversarial Attacks on Deep Learning Wi-Fi Fingerprinting. (arXiv:2301.03760v1 [cs.CR])

Empowered by deep neural networks (DNNs), Wi-Fi fingerprinting has recently
achieved astonishing localization performance to facilitate many
security-critical applications in wireless networks, but it is inevitably
exposed to adversarial attacks, where subtle perturbations can mislead DNNs to
wrong predictions. Such vulnerability provides new security breaches to
malicious devices for hampering wireless network security, such as
malfunctioning geofencing or asset management. The prior adversarial attack on
localization DNNs uses additive perturbations on channel state information
(CSI) measurements, which is impractical in …

adversarial adversarial attacks applications asset asset management attack attacks breaches critical deep learning devices empowered exposed fingerprinting geofencing localization malicious management network networks network security neural networks performance predictions security security breaches vulnerability wi-fi wireless