1

We propose a taxonomy to categorize and better understand black-box attacks revealing unexplored threat spaces and other interesting findings, and emphasize the need to consider resource costs like attack runtime to evaluating attacks.

We propose a taxonomy to categorize and better understand black-box attacks revealing unexplored threat spaces and other interesting findings, and emphasize the need to consider resource costs like attack runtime to evaluating attacks.

We demonstrated that some data distributions can be inherently robust to poisoning, and that improving distributional quality can enhance resistance to poisoning attacks.

We demonstrated that some data distributions can be inherently robust to poisoning, and that improving distributional quality can enhance resistance to poisoning attacks.

We introduce a method to manipulate neuron activations while pre-training models, allowing highly successful inference of sensitive properties of the victim’s downstream training data.

We introduce a method to manipulate neuron activations while pre-training models, allowing highly successful inference of sensitive properties of the victim’s downstream training data.

We propose efficient data poisoning attacks that can asymptotically approach a target model with desired properties.

We propose efficient data poisoning attacks that can asymptotically approach a target model with desired properties.

We propose query-efficient black-box attacks that innovatively combine existing methods and prioritize targeting more vulnerable seeds.

We propose query-efficient black-box attacks that innovatively combine existing methods and prioritize targeting more vulnerable seeds.