โ๏ธ Knowledge Editing¶
๐ฌ EMNLP2026 ยท 4 paper notes
๐ Same area in other venues: ๐ง NeurIPS2026 (8) ยท ๐ท CVPR2026 (2) ยท ๐ฌ ICLR2026 (15) ยท ๐ฌ ACL2026 (10) ยท ๐งช ICML2026 (8) ยท ๐ค AAAI2026 (4)
- Editing Rules in LLMs Requires More Than a Localized Update
-
Addressing the failure of localized factual model editors on mathematical and physical rules, this paper reveals via fine-grained causal tracing that rule knowledge exhibits form-specific layer-wise separation (formulas and descriptions reside in early layers while numerical instances peak in middle layers), and proposes Distributed Multi-Layer Editing (DMLE) to perform coordinated cross-layer updates that dramatically boost rule understanding and instance portability.
- Generalizable Lifelong Model Editing via Preference Optimization
-
To tackle superficial prompt overfitting and catastrophic forgetting in lifelong knowledge editing, GLIME formulates continual editing as a joint optimization problem combining factual injection, edit-independent preference optimization for behavior-level regularization, experience replay, and randomized orthogonal gradient projection, achieving superior generalization and downstream portability under realistic autoregressive decoding.
- Measuring the Depth of LLM Unlearning via Activation Steering
-
Proposes the Unlearning Depth Score (UDS), a training-free, causal, and dataset-invariant metric that evaluates the recoverability of forgotten knowledge across internal layers via two-stage activation patching, outperforming observational probes and output-level metrics in detecting residual knowledge obscured by representation shifts.
- Validator-Guided Decoding for Multi-Hop Knowledge Editing
-
To resolve the failure of edited language models to compose updated facts in multi-hop reasoning despite retaining strong fact-checking capabilitiesโtermed the generate-verification gapโthis paper proposes Validator-Guided Decoding (VGD), which employs a lightweight trained validator as an external correctness signal to early-commit high-confidence candidates and selectively revise inconsistent predictions at test time.