Adam Dziedzic

We introduce a bitwise, radioactive watermark for bitwise image generative models and show that it remains robust against various attacks while preserving high image quality and generation speed.

We propose the first label memorization framework for node classification in GNNs and investigate the relationship between memorization and graph/node properties.

We perform large-scale strong MIAs on pre-trained LLMs to clarify the extent of actual privacy risk MIAs pose in this setting.