Backdoor learning

The paper highlights the importance of detecting dirty samples in datasets used for Data-centric AI, such as poisoned samples or noisy labels, which compromise the reliability of DNNs. To address this, we propose the Versatile Data Cleanser (VDC), a novel approach leveraging multimodal large language models to identify inconsistencies between images and their labels, demonstrating its effectiveness across various types of dirty samples.

Adversarial phenomenon has been widely observed in machine learning (ML) systems, especially in those using deep neural networks, …

Deep neural networks (DNNs) can be manipulated to exhibit specific behaviors when exposed to specific trigger patterns, without …

Data-poisoning based backdoor attacks aim to insert backdoor into models by manipulating training datasets without controlling the …

Adversarial machine learning (AML) studies the adversarial phenomenon of machine learning, which may make inconsistent or unexpected …

BackdoorBench aims to provide an easy implementation of both backdoor attack and backdoor defense methods to facilitate future research, as well as a comprehensive evaluation of existing attack and defense methods. We further present analysis from different perspectives about these 8,000 evaluations, studying the effects of attack against defense algorithms, poisoning ratio, model and dataset in backdoor learning.