Deep Researcher with Test-Time Diffusion

2025-07-21

AI Summary

Existing deep research agents exhibit performance saturation when generating lengthy, complex reports, primarily due to reliance on generic test-time scaling. To address this, we propose a draft-guided diffusion framework for research synthesis. Our method models report generation as an iterative optimization process: an initial draft anchors the research direction; dynamic retrieval augmentation and multi-hop reasoning enable continuous information injection; and a self-evolving mechanism coordinates the co-adaptation of workflow components—including search, reasoning, and revision. Integrating test-time diffusion, context-aware retrieval, and learnable revision policies, our approach significantly improves factual accuracy, logical coherence, and temporal relevance of generated reports. Evaluated on multiple benchmarks requiring intensive search and deep reasoning, our method achieves state-of-the-art performance, outperforming existing deep research agents by an average of 12.7% in report quality score.

Technology Category

• Search and Optimization: Learning to Search
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Deep research agents, powered by Large Language Models (LLMs), are rapidly advancing; yet, their performance often plateaus when generating complex, long-form research reports using generic test-time scaling algorithms. Drawing inspiration from the iterative nature of human research, which involves cycles of searching, reasoning, and revision, we propose the Test-Time Diffusion Deep Researcher (TTD-DR). This novel framework conceptualizes research report generation as a diffusion process. TTD-DR initiates this process with a preliminary draft, an updatable skeleton that serves as an evolving foundation to guide the research direction. The draft is then iteratively refined through a "denoising" process, which is dynamically informed by a retrieval mechanism that incorporates external information at each step. The core process is further enhanced by a self-evolutionary algorithm applied to each component of the agentic workflow, ensuring the generation of high-quality context for the diffusion process. This draft-centric design makes the report writing process more timely and coherent while reducing information loss during the iterative search process. We demonstrate that our TTD-DR achieves state-of-the-art results on a wide array of benchmarks that require intensive search and multi-hop reasoning, significantly outperforming existing deep research agents.

Problem

• Improving long-form research report generation using diffusion
• Enhancing iterative refinement with dynamic retrieval mechanisms
• Achieving state-of-the-art performance in multi-hop reasoning tasks

Innovation

• Diffusion process for research report generation
• Iterative refinement with dynamic retrieval mechanism
• Self-evolutionary algorithm enhancing agentic workflow