WARP: Weight Teleportation for Attack-Resilient Unlearning Protocols

Researchers introduce WARP, a new defense mechanism for machine unlearning protocols that protects against privacy attacks where adversaries can exploit differences between pre- and post-unlearning AI models. The technique reduces attack success rates by up to 92% while maintaining model accuracy on retained data.

• →Current machine unlearning methods are vulnerable to membership inference and data reconstruction attacks that exploit model parameter differences.
• →WARP uses neural network symmetries to obfuscate forgotten data signals through weight teleportation and parameter dispersion.
• →The defense achieves up to 64% reduction in black-box attacks and 92% in white-box attacks across six unlearning algorithms.
• →The approach works as a plug-and-play solution that can be applied to existing state-of-the-art unlearning methods.
• →Results demonstrate teleportation as a general privacy protection tool for approximate machine unlearning systems.