Self-supervised learning (SSL) is vulnerable to backdoor attacks, while the downstream classifiers based on SSL models inevitably inherit these backdoors, even when they are trained on clean samples. Despite the proposal of several methods of backdoor defense against backdoor attacks, few methods remain that can be used to effectively defend against various backdoor attacks while maintaining the high performance of the model. In this paper, based on the discovery that unlearning any trigger enhances the overall backdoor robustness of the model, a novel, efficient, and straightforward approach is proposed to address the most advanced backdoor attacks. This method involves two stages. Firstly, a backdoor is actively implanted in the model with a custom trigger. Secondly, the model is fine-tuned to unlearn the custom trigger. Through these two stages, it is found out that not only is the implanted backdoor removed, but the unknown backdoors implanted by attackers are also effectively mitigated. As illustrated by the extensive experiments conducted on multiple datasets against the current state-of-the-art methods of attack, the method proposed in this paper requires only a small amount of clean data (approximately 1%) to reduce the success rate of backdoor attacks effectively while ensuring minimal impact on model performance.

Self-supervised learning (SSL) has gained significant attention for its ability to leverage large-scale unlabeled data for training, demonstrating remarkable applications in computer vision and natural language processing. However, recent studies have shown that SSL models are highly vulnerable to backdoor attacks, where adversaries can implant hidden triggers into the pretraining process, leading to compromised downstream tasks. Existing backdoor defense mechanisms, while effective to some extent, often rely on labeled data, incur high computational costs, or involve complex operations, making it difficult to achieve a balance between model performance and defense effectiveness. In this paper, we propose a backdoor defense method called UTRBD, which is based on untargeted trigger reconstruction, combined with a contrastive fine-tuning strategy. It effectively defends against backdoor attacks using only a small amount of unlabeled data and minimal computational resources, while maintaining high model performance. Through extensive experiments on the ImageNet100 and CIFAR10 datasets, our method performs excellently across various self-supervised learning models and different encoder architectures, successfully mitigating multiple types of backdoor attacks.