Research map / Large language models / Retrieval and knowledge
ActiShade: Activating Overshadowed Knowledge to Guide Multi-Hop Reasoning in Large Language Models
Huipeng Ma, Luan Zhang, Dandan Song, Linmei Hu, Yuhang Tian, Jun Yang, Changzhi Zhou, Chenhao Li, Yizhou Jin, Xudong Li, Meng Lin, Mingxing Zhang and 1 more
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 Knowledge Overshadowing Detection (GaP) to multi-round ActiShade (ablation) | ActiShade (Llama-3-8B) · MuSiQue | F1: 22.83 -> 26.94 | Improved |
| add GaP detection to ActiShade (ablation) | ActiShade (Llama-3-8B) · HotpotQA | F1: 51.23 -> 56.33 | Improved |
| add GaP detection to ActiShade (ablation) | ActiShade (Llama-3-8B) · 2WikiMQA | F1: 45.18 -> 46.02 | Improved |
| add GaP detection to single-round retrieval (ablation) | Direct-R (Llama-3-8B) · 2WikiMQA | F1: 35.56 -> 38.67 | Improved |
| replace GaP by CoDA token-removal detection (ablation) | ActiShade (Llama-3-8B) · 2WikiMQA | F1: 46.02 -> 41.29 | Hurt |
| add CoDA detection to ActiShade without detection (ablation) | ActiShade-NoKOD (Llama-3-8B) · MuSiQue | F1: 22.83 -> 21.23 | Hurt |
| add CoDA detection to single-round retrieval (ablation) | Direct-R (Llama-3-8B) · HotpotQA | F1: 44.89 -> 45.98 | Improved |
| train retriever with fine-grained contrastive loss (positive/semi-positive/negative) instead of standard contrastive (ablation) | ActiShade retriever · MuSiQue | F1: 24.10 -> 26.94 | Improved |
| add fine-grained contrastive retriever fine-tuning vs base retriever (ablation) | contriever-msmarco · Retriever test, positive docs | Recall@1: 29.20 -> 75.33 | Improved |
| fine-grained instead of standard contrastive loss for positive plus semi-positive docs at Recall@1 (ablation) | ActiShade retriever · Retriever test, Pos&Semi | Recall@1: 38.21 -> 38.14 | No clear effect |
| vary Gaussian noise sigma from 0.05 to 0.5 in GaP (ablation) | ActiShade (Llama-3-8B) · multi-hop QA | F1 | Mixed |
| scale generator from Qwen2.5-7B to 14B | ActiShade · HotpotQA | F1: 55.45 -> 57.45 | Improved |
Closest papers by meaning
- Retrieval is Not Enough: Enhancing RAG through Test-Time Critique and Optimization (NeurIPS 2025 · similarity 0.88)
- Simple is Effective: The Roles of Graphs and Large Language Models in Knowledge-Graph-Based Retrieval-Augmented Generation (ICLR 2025 · similarity 0.88)
- Adapting to Non-Stationary Environments: Multi-Armed Bandit Enhanced Retrieval-Augmented Generation on Knowledge Graphs (AAAI 2025 · similarity 0.87)
- Making Retrieval-Augmented Language Models Robust to Irrelevant Context (ICLR 2024 · similarity 0.87)
- Information Re-Organization Improves Reasoning in Large Language Models (NeurIPS 2024 · similarity 0.87)
- From Complex to Atomic: Enhancing Augmented Generation via Knowledge-Aware Dual Rewriting and Reasoning (ICML 2025 · similarity 0.87)
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.
