Retain Cross-Leverage
A type of leverage matrix that captures the interaction effects between forgotten samples and retained samples, used in machine unlearning to analyze the structure of the output-space gap.
Timeline (last 90 days)
Two leverage matrices emerge in the framework: forget self-leverage and retain cross-leverage, capturing the entanglement of forget and retain data, respectively, under the geometry of the retrained model.
These two leverage matrices are shown both theoretically and experimentally to drive the influence of the unlearning residual on forget outputs and retain outputs.
All Facts (2)
Two leverage matrices emerge in the framework: forget self-leverage and retain cross-leverage, capturing the entanglement of forget and retain data, respectively, under the geometry of the retrained model.
50%UnverifiedThese two leverage matrices are shown both theoretically and experimentally to drive the influence of the unlearning residual on forget outputs and retain outputs.
50%