DOW-KE: Anchor-Free Multi-Layer Knowledge Editing via Direct End-to-End Weight Optimization
DOW-KE backpropagates the final editing objective through the complete model, jointly optimizing the updates of all edited layers so cross-layer propagation and coupling enter every gradient step, and achieves the highest overall Score and neighborhood Specificity among the evaluated baselines.
Ran Chen, Junbo Zhang, Qianli Zhou et al.
· 0 citations