Research map / Image generation / Fast diffusion sampling
MAnchors: Memorization-Based Acceleration of Anchors via Rule Reuse and Transformation
Haonan Yu, Junhao Liu, Xin Zhang
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 |
|---|---|---|---|
| add memorization with horizontal and vertical rule transformation | Anchors · Sentiment Analysis (Llama) | average time (s): 513.06 -> 58.70 (8.74×) | Improved |
| add memorization with rule transformation | Anchors · Income Prediction (RF) | average time (s): 8.51 -> 4.02 (2.11×) | Improved |
| add memorization with rule transformation | Anchors · Sentiment Analysis (RF) | average time (s): 32.10 -> 18.92 (1.69×) | Improved |
| add memorization with rule transformation | Anchors · Image Classification (YOLOv8) | average time (s): 61.34 -> 31.91 (1.92×) | Improved |
| add memorization with rule transformation | Anchors · SST-2 (Gemma-2-2B) | average time (s): 269.96 s to 59.17 s | Improved |
| add memorization with rule transformation | Anchors · AG News (Gemma-2-2B) | average time (s): 259.31 s to 88.96 s | Improved |
| add memorization with rule transformation | Anchors · Sentiment Analysis (Llama) | coverage: 0.10 -> 0.09 | Hurt |
| add memorization with rule transformation | Anchors · Sentiment Analysis (Llama) | rule length: 4.34 -> 5.39 | Hurt |
| add memorization with rule transformation | Anchors · SST-2 (Gemma-2-2B) | coverage: 0.17 to 0.13 | Hurt |
| add memorization with rule transformation | Anchors · AG News (Gemma-2-2B) | coverage: 0.50 to 0.48 | No clear effect |
| add memorization with rule transformation | Anchors · Memory usage | average memory (MB): 276.81 -> 279.97 | Hurt |
| increase similarity threshold tau_sim (ablation) | MAnchors · RF, NN, GBT parameter study | coverage | Improved |
Closest papers by meaning
- Memorize Theorems, Not Instances: Probing SFT Generalization through Mathematical Reasoning (NeurIPS 2026 · similarity 0.83)
- Accelerating the Global Aggregation of Local Explanations (AAAI 2024 · similarity 0.83)
- Frozen Memory Is Not Enough: Rethinking External Memory as Extraction (NeurIPS 2026 · similarity 0.83)
- ReX: A Framework for Incorporating Temporal Information in Model-Agnostic Local Explanation Techniques (AAAI 2025 · similarity 0.83)
- Memory is Reconstructed, Not Retrieved: Graph Memory for LLM Agents (ICML 2026 · similarity 0.83)
- Towards Long-Horizon Interpretability: Efficient and Faithful Multi-Token Attribution for Reasoning LLMs (ICML 2026 · similarity 0.83)
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.
