Backdoor Attacks Optimized through Genetic Algorithm-Driven Data Augmentation Combinations in Deep Neural Networks

Deep neural networks (DNNs) have become increasingly popular across a wide range of applications, but recent research has revealed their vulnerability to backdoor attacks. However, existing backdoor attack methods often involve complex computational costs, making them less practical in real-world scenarios. Data augmentation, a commonly used technique for improving model generalization, offers an ideal source for trigger generation due to its low computational overhead and the fact that the generated images do not raise suspicion. However, current backdoor attack strategies that employ data augmentation typically rely on randomly selected augmentation techniques, which yield suboptimal results. In this paper, we propose BGD3A, a novel backdoor attack method driven by a genetic algorithm, optimizing the selection of data augmentation combinations. Attackers can customize the fitness function to generate an optimal combination of augmentation techniques to meet their objectives. Extensive experimental results demonstrate that BGD3A not only achieves high attack success rates but also remains robust against state-of-the-art backdoor defense mechanisms.