Technique stabilizes hidden attacks in shared AI image models

Original: Awakening the Hydra: Stabilizing Multi-Concept Backdoor Injection in Text-to-Image Diffusion Models

Source: arxiv.org ↗