Sponge Examples: Energy-Latency Attacks on Neural Networks

TL;DR

This paper introduces sponge examples, a new type of adversarial input that significantly increases neural network energy consumption and latency, posing a threat to real-time systems and hardware efficiency.

Contribution

It presents the concept of sponge examples, demonstrates their effectiveness across multiple models and hardware, and proposes a defense strategy to mitigate these attacks.

Findings

Energy consumption increased by 10 to 200 times

Attacks effective across CPUs, GPUs, and ASICs

Potential to delay real-time neural network decisions

Abstract

The high energy costs of neural network training and inference led to the use of acceleration hardware such as GPUs and TPUs. While this enabled us to train large-scale neural networks in datacenters and deploy them on edge devices, the focus so far is on average-case performance. In this work, we introduce a novel threat vector against neural networks whose energy consumption or decision latency are critical. We show how adversaries can exploit carefully crafted $\boldsymbol{sponge}~\boldsymbol{examples}$, which are inputs designed to maximise energy consumption and latency. We mount two variants of this attack on established vision and language models, increasing energy consumption by a factor of 10 to 200. Our attacks can also be used to delay decisions where a network has critical real-time performance, such as in perception for autonomous vehicles. We demonstrate the portability of our malicious inputs across CPUs and a variety of hardware accelerator chips including GPUs, and an ASIC simulator. We conclude by proposing a defense strategy which mitigates our attack by shifting the analysis of energy consumption in hardware from an average-case to a worst-case perspective.