SAAS: Self-Aware Reinforcement Learning for Over-Search Mitigation in Agentic Search
Summary
SAAS introduces a reinforcement learning framework that enhances agent self-awareness to reduce unnecessary searches in LLM-based question answering systems, balancing accuracy and computational cost.
View Cached Full Text
Cached at: 06/01/26, 11:20 AM
Paper page - SAAS: Self-Aware Reinforcement Learning for Over-Search Mitigation in Agentic Search
Source: https://huggingface.co/papers/2605.29796
Abstract
SAAS introduces a reinforcement learning framework that enhances agent self-awareness to reduce unnecessary searches in LLM-based question answering systems.
Agentic searchenablesLLMsto solve complexmulti-hop questionsthroughiterative reasoningandexternal search. Despite the effectiveness, these systems often suffer from a critical limitation in practice: agents fail to recognize their own knowledge boundaries, blindly triggering searches when internal knowledge suffices and failing to terminate search even when adequate evidence has been collected. The lack ofself-awarenessleads to severeover-search, incurring substantial inference latency and prohibitive computational cost. To this end, we propose SAAS, a novelRL frameworkdesigned to cultivate dynamicself-awarenessthat precisely regulates search behavior without compromising accuracy. SAAS introduces three key components: (i) asearch boundary modelingmechanism, which identifies the search boundary under the evolving policy by contrasting search-disabled and search-enabled rollouts; (ii) aboundary-aware reward module, which translates this boundary awareness intotrajectory-level penalties, suppressing unnecessary and redundant searches; and (iii) astage-wise optimizationstrategy, which leverages a sequential curriculum to prioritize reasoning over search regularization, thereby avoidingreward hacking. Extensive experiments demonstrate that SAAS substantially reducesover-search, while maintaining accuracy. Our code is anonymously released at https://github.com/XMUDeepLIT/SAAS.
View arXiv pageView PDFGitHub5Add to collection
Get this paper in your agent:
hf papers read 2605\.29796
Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash
Models citing this paper0
No model linking this paper
Cite arxiv.org/abs/2605.29796 in a model README.md to link it from this page.
Datasets citing this paper0
No dataset linking this paper
Cite arxiv.org/abs/2605.29796 in a dataset README.md to link it from this page.
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2605.29796 in a Space README.md to link it from this page.
Collections including this paper0
No Collection including this paper
Add this paper to acollectionto link it from this page.
Similar Articles
SlimSearcher: Training Efficiency-Aware Web Agents via Adaptive Reward Gating
SlimSearcher is a framework that improves efficiency in deep research agents by combining Pareto-efficient trajectory filtering and adaptive reward shaping, reducing tool-call rounds by 17-58% while maintaining accuracy on benchmarks like GAIA, BrowseComp, and XBenchDeepSearch.
Learning to Adapt: Self-Improving Web Agent via Cognitive-Aware Exploration
Proposes SCALE, a framework for self-improving web agents using cognitive-aware exploration with three adversarial roles and a graph exploration strategy. Also introduces a large-scale dataset SCALE-20k from real websites, showing significant improvements in MLLM-based web agents.
GRASP: GRanularity-Aware Search Policy for Agentic RAG
Introduces GRASP, a reinforcement learning framework that trains agents to adaptively coordinate semantic search, keyword search, and paragraph reading during multi-step reasoning, improving retrieval recall and question answering performance on multi-hop benchmarks.
Scaling Retrieval-Augmented Reasoning with Parallel Search and Explicit Merging
Introduces MultiSearch, an RL-based framework that generates multiple queries at each reasoning step and explicitly merges retrieved information to improve signal-to-noise ratio and reasoning accuracy in question-answering tasks.
SAGE: Scalable Automated Robustness Augmentation for LLM Knowledge Evaluation
This paper introduces SAGE, a framework for scalable automated robustness augmentation of LLM knowledge evaluation benchmarks. It uses fine-tuned smaller models with reinforcement learning to generate and verify question variants at a lower cost than existing methods.