Research map / Large language models / Retrieval and knowledge
ECD: Evidence-guided Contrastive Decoding in Retrieval-Augmented Generation with Accurate Knowledge Reference Adjustment
Yize Sui, Yan Xu, Kun Hu, Jing Ren, Wenjing Yang
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 |
|---|---|---|---|
| use dynamic evidence-based weight instead of fixed w_ECD = 0.6 (ablation) | ECD (LLaMA-3-8B-Instruct) · PopQA | EM: 42.73 -> 43.59 | Improved |
| use dynamic evidence-based weight instead of fixed w_ECD = 0.6 (ablation) | ECD (LLaMA-3-8B-Instruct) · HotpotQA | EM: 42.86 -> 47.23 | Improved |
| use dynamic weight instead of fixed w_ECD = 0.4 (ablation) | ECD (LLaMA-3-8B-Instruct) · TriviaQA | EM: 62.63 -> 67.89 | Improved |
| add retrieved context (w_ECD 0 to 1, RegCls to RegOpn) (ablation) | ECD (LLaMA-3-8B-Instruct) · PopQA | EM: 30.36 -> 40.03 | Improved |
| use evidence-based ECD instead of entropy-based ACD with noisy context | ECD (LLaMA-3-8B-Instruct) · TriviaQA noisy context | EM: 56.38 -> 62.30 | Improved |
| use ECD instead of regular open-book decoding with noisy context | ECD (LLaMA-3-8B-Instruct) · TriviaQA noisy context | EM: 37.55 -> 62.30 | Improved |
| use ECD instead of RegOpn with gold context | ECD (LLaMA-3-8B-Instruct) · TriviaQA gold context | EM: 84.25 -> 85.36 | Improved |
| use ECD instead of second-best method on average | ECD (LLaMA-3-8B-Instruct) · four-dataset average | EM: 49.58 -> 51.89 | Improved |
| use ECD instead of CDA on average | ECD (LLaMA-3-8B-Instruct) · four-dataset average | EM: 47.33 -> 51.89 | Improved |
| use static context emphasis CAD instead of regular open-book decoding | CAD (LLaMA-3-8B-Instruct) · four-dataset average | EM: 47.26 -> 44.66 | Hurt |
| increase top-K candidate tokens from 1 to 25 (ablation) | ECD (LLaMA-2-7B-Chat) · TriviaQA | EM: about 57 -> approximately 65 | Improved |
| increase top-K candidate tokens from 25 to 50 (ablation) | ECD (LLaMA-3-8B-Instruct) · PopQA | EM: roughly 43 -> about 44 | No clear effect |
Closest papers by meaning
- Chain-of-Retrieval Augmented Generation (NeurIPS 2025 · similarity 0.86)
- Speculative RAG: Enhancing Retrieval Augmented Generation through Drafting (ICLR 2025 · similarity 0.86)
- RAPID: Long-Context Inference with Retrieval-Augmented Speculative Decoding (ICML 2025 · similarity 0.86)
- A Theory for Token-Level Harmonization in Retrieval-Augmented Generation (ICLR 2025 · similarity 0.86)
- Long-Context LLMs Meet RAG: Overcoming Challenges for Long Inputs in RAG (ICLR 2025 · similarity 0.86)
- Retrieval from Within: An Intrinsic Capability of Attention-Based Models (NeurIPS 2026 · similarity 0.86)
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.
