Reinforcement Learning Inspired Black-box Adversarial Attacks for Computer Vision
A reinforcement-learning black-box attack fools vision models with fewer queries than earlier methods.
The paper proposes RIBA, a black-box adversarial attack that borrows reinforcement-learning ideas to optimize perturbations against non-differentiable computer-vision models. Compared with prior attacks, it reports 25.4% fewer median queries to attack a ResNet-18 on CIFAR-10 and 22.5% fewer to fool a ViT-B/16 on ImageNet. The authors also say RIBA can match white-box attack performance against an adversarially trained model. No real-world incident is described.
- RIBA optimizes black-box perturbations using reinforcement-learning ideas.
- It used 25.4% fewer median queries against ResNet-18 on CIFAR-10.
- It used 22.5% fewer median queries against ViT-B/16 on ImageNet.
- Authors say it can match white-box results on an adversarially trained model.
Full article213 words · extracted from arxiv.org · click to collapse
Neural networks, both convolution or transformer based, are essential for modern computer vision systems. However, they are vulnerable to small perturbations, almost imperceptible to humans, which significantly alter the model's prediction. These adversarial attacks are often considered to be a significant threat to the implementation of neural networks in safety-critical applications. Most attacks utilize the white-box threat model and therefore require full access to the target model, making them unrealistic to use in practice. We propose a novel approach under the more realistic black-box threat model that utilizes concepts from reinforcement learning to optimize perturbations with a non-differentiable target model. Reinforcement learning algorithms have already been optimized to be query efficient, making them an ideal starting point when designing black-box adversarial attacks. We show the success of our reinforcement learning inspired black-box adversarial attack (RIBA) in generating adversarial perturbations using only a small number of queries to the target model, by comparing it to state of the art attacks on different models on the Cifar10 and ImageNet data sets. RIBA takes $25.4\%$ fewer median queries to generate attacked images against a ResNet-18 on Cifar10 and $22.5\%$ fewer median queries to fool a Vit-B/16 model on ImageNet. Additionally, we demonstrate that RIBA can match the performance of white-box attacks on an adversarially trained model.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2609.24249