Machine unlearning (MU) empowers individuals with the'right to be forgotten' by removing their private or sensitive information encoded in machine learning models. However, it remains uncertain whether MU can be effectively applied to Multimodal Large Language Models (MLLMs), particularly in scenarios of forgetting the leaked visual data of concepts.

With respect to experiments, our contributions are twofold.

Vision-language (VL) pre-training has recently received considerable attention.

As aresult, there is aproliferation of distinct architectures and loss functions for different vision tasks.

Moreover,despiteanonymization efforts, privacy risks have been associated with publicly released user data [38].

Conditional image synthesis aims tocreate an image according tothe given control signals.

In this work, we introduce WinoGAViL: an online game of vision-andlanguage associations (e.g., betweenwerewolves and a full moon), used as a dynamic evaluation benchmark.

Then in the stage 2, an auto-regressive model (such as PixelCNN [47]) learns to fit the prior of hidden variables.