ToxicTextCLIP: Text-Based Poisoning and Backdoor Attacks on CLIP Pre-training
NegativeArtificial Intelligence
The CLIP model has made great strides in vision-language tasks, but its dependence on unfiltered internet data raises serious concerns about data poisoning and backdoor attacks. While most research has focused on image-related vulnerabilities, the risks associated with text data are equally significant and warrant attention.
— Curated by the World Pulse Now AI Editorial System


