Interference-Aware Multi-Task Unlearning

In three linesNew paper on multi-task machine unlearning: removing training data from shared models without degrading other tasks. Proposes interference-aware framework combining task-aware gradient projection and instance-level orthogonalization. Reduces interference by 30–53% on vision benchmarks.

Summary generated by Claude — human-verified