Research map / Large language models / Retrieval and knowledge
Optimizing Agentic Reasoning with Retrieval via Synthetic Semantic Information Gain Reward
Senkang Hu, Yong Dai, Yuzhi Zhao, Yihang Tao, Yu Guo, Zhengru Fang, Sam Kwong, Yuguang Fang
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 information gain reward (lambda=0.6 vs 0.0) (ablation) | InfoReasoner-3B · Seven QA benchmarks average | exact match: 29.9% -> 34.6% | Improved |
| raise lambda from 0.6 to 0.8 (ablation) | InfoReasoner-3B · Seven QA benchmarks average | exact match: 34.6% -> 30.4% | Hurt |
| raise lambda from 0.6 to 1.0 (ablation) | InfoReasoner-3B · Seven QA benchmarks average | exact match: 34.6% -> 32.0% | Hurt |
| lower lambda from 0.6 to 0.4 (ablation) | InfoReasoner-3B · Seven QA benchmarks average | exact match: 0.346 -> 0.315 | Hurt |
| lower lambda from 0.6 to 0.2 (ablation) | InfoReasoner-3B · Seven QA benchmarks average | exact match: 0.346 -> 0.327 | Hurt |
| lambda=0.6 vs 0.0 on TriviaQA (ablation) | InfoReasoner-3B · TriviaQA | exact match: 0.538 -> 0.634 | Improved |
| lambda=0.6 vs 0.0 on 2WikiMultiHopQA (ablation) | InfoReasoner-3B · 2WikiMultiHopQA | exact match: 0.254 -> 0.324 | Improved |
| lambda=0.6 vs 0.0 on Bamboogle (ablation) | InfoReasoner-3B · Bamboogle | exact match: 0.118 -> 0.144 | Improved |
| InfoReasoner-3B vs Search-R1-3B-Instruct | InfoReasoner · Seven QA benchmarks average | exact match: 30.1% -> 34.6% | Improved |
| InfoReasoner-7B vs AutoRefine-7B-Base | InfoReasoner · Seven QA benchmarks average | exact match: 36.9% -> 39.1% | Improved |
| InfoReasoner-3B vs Search-R1-3B-Instruct on Bamboogle | InfoReasoner-3B · Bamboogle | exact match: 0.184 -> 0.144 | Hurt |
| InfoReasoner-7B vs ZeroTIR-7B on tool-integrated math | InfoReasoner-7B · MATH500 | accuracy: 80.2% -> 85.8% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SimpleTIR-7B — SimpleTIR: End-to-End Reinforcement Learning for Multi-Turn Tool-Integrated Reasoning (ICLR 2026) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Iterative Self-Incentivization Empowers Large Language Models as Agentic Searchers (NeurIPS 2025 · similarity 0.89)
- MC-Search: Evaluating and Enhancing Multimodal Agentic Search with Structured Long Reasoning Chains (ICLR 2026 · similarity 0.88)
- Scent of Knowledge: Optimizing Search-Enhanced Reasoning with Information Foraging (NeurIPS 2025 · similarity 0.88)
- Empowering Efficiency and Efficacy in WebAgent via Enabling Info-Rich Seeking (ICLR 2026 · similarity 0.88)
- Mind2Web 2: Evaluating Agentic Search with Agent-as-a-Judge (NeurIPS 2025 · similarity 0.88)
- ReSeek: A Self-Correcting Framework for Search Agents with Instructive Rewards (ICML 2026 · similarity 0.88)
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.
