🚨 Excited to share that our work, Robust Adversarial Concept Erasure for #T2I (R.A.C.E.), has been accepted for an ORAL presentation at ECCV (@eccvconf)!
Please check: https://t.co/ubqWUMf44Y
Thank you to @kyleminbs and @prof_yz.
Working on generative models? Attending #ECCV? You should check out our tutorial - Responsibly Building Generative Models.
⏰ - 9/29 (Sunday), 2 PM
📍 - Suite 7 (In-person)
💻 - https://t.co/oc4ZBy3KCL
We will discuss different aspects of failures in #GenAI !
#tutorial #computervision
🚨Introducing R.A.C.E.: a novel adversarial training approach to fortify concept erasure without extra modules.
🔹Single-timestep Adv. attack
🔹Defends against white/black-box prompt attacks
🔹Enhances model robustness
Read more: https://t.co/ubqWUMfBUw
@kyleminbs, @prof_yz