Sponge Attack

Sponge attacks are a class of adversarial attacks targeting deep learning models, aiming to increase their energy consumption and inference latency without significantly degrading their accuracy. Current research focuses on developing and analyzing these attacks across various model architectures, including generative adversarial networks (GANs) and autoencoders, and exploring their effectiveness in both on-device and server settings, including federated learning scenarios. This line of research highlights the vulnerability of resource-constrained systems, such as mobile devices, to these attacks and underscores the need for robust defense mechanisms to mitigate their impact on energy efficiency and performance.