Research map / Architectures and efficiency / RNNs and normalization
From Backward Spreading to Forward Replay: Revisiting Target Construction in LLM Parameter Editing
Wei Liu, Liu Hongkai, Zhiying Deng, Yee Whye Teh, Wee Sun Lee
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace backward spreading with forward replay (FE) | MEMIT, Llama3-8B-Instruct · MCF | efficacy accuracy: 94.3 -> 99.8 | Improved |
| replace backward spreading with forward replay (FE) | MEMIT, Llama3-8B-Instruct · ZsRE | efficacy accuracy: 82.0 -> 93.0 | Improved |
| replace backward spreading with forward replay (FE) | MEMIT, Llama3-8B-Instruct · ZsRE | D_KL: 0.60 -> 0.09 | Improved |
| add forward replay to AlphaEdit | AlphaEdit, Llama3-8B-Instruct · MCF | efficacy accuracy: 89.0 -> 99.2 | Improved |
| add forward replay to PRUNE | PRUNE, Llama3-8B-Instruct · ZsRE | generalization accuracy: 72.7 -> 84.4 | Improved |
| add forward replay to RECT | RECT, Llama3-8B-Instruct · MCF | generalization accuracy: 46.7 -> 60.4 | Improved |
| add forward replay to AlphaEdit | AlphaEdit, Llama3-8B-Instruct · MCF | D_KL: 0.63 -> 0.47 | Improved |
| edit multiple layers instead of only the last layer (ablation) | MEMIT, Llama3-8B-Instruct · MCF | efficacy accuracy: 59.7 -> 94.3 | Improved |
| add forward replay instead of backward spreading for remaining residual (ablation) | MEMIT, Llama3-8B-Instruct · Layer8 remaining residual ratio | remaining residual ratio: 0.36 -> 0.03 | Improved |
| remove division of residual by remaining layer count (No-dividing) (ablation) | MEMIT backward spreading · Layer4 remaining residual ratio | remaining residual ratio: 0.96 -> 1.19 | Hurt |
| remove division of residual by remaining layer count (No-dividing) (ablation) | MEMIT backward spreading · Layer8 remaining residual ratio | remaining residual ratio: 0.36 -> 0.23 | Improved |
| use FE instead of BLUE (independent targets) | BLUE, Llama3-8B-Instruct · ZsRE | generalization accuracy: 78.2 -> 84.3 | Improved |
Closest papers by meaning
- Rethinking Residual Distribution in Locate-then-Edit Model Editing (NeurIPS 2025 · similarity 0.85)
- PnP Inversion: Boosting Diffusion-based Editing with 3 Lines of Code (ICLR 2024 · similarity 0.84)
- h-Edit: Effective and Flexible Diffusion-Based Editing via Doob's h-Transform (CVPR 2025 · similarity 0.84)
- Norm Anchors Make Model Edits Last (NeurIPS 2026 · similarity 0.84)
- Meta-CoT: Enhancing Granularity and Generalization in Image Editing (CVPR 2026 · similarity 0.84)
- Inversion-Free Image Editing with Language-Guided Diffusion Models (CVPR 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
