Agentic-Ideation: Sample Efficient Agentic Trajectories Synthesis for Scientific Ideation Agents

2026-06-30

AI Summary

Existing approaches to scientific idea generation are constrained by fixed agent workflows, which struggle to efficiently explore the vast literature and reasoning space while incurring high data synthesis costs. This work proposes an efficient agent trajectory synthesis framework that constructs a tool space integrating external and cognitive tools and introduces an Oracle-guided multi-agent trajectory synthesis mechanism. By transforming aimless trial-and-error into directed generation, the method substantially improves sample efficiency for high-quality trajectories. Additionally, a tool execution result masking strategy is employed during training to strengthen the model’s decision-making logic. Experimental results demonstrate that the proposed approach outperforms state-of-the-art workflow baselines by 11.91% in idea quality and achieves over a tenfold improvement in sample efficiency.

Abstract

Ideation plays a pivotal role in scientific discovery. Recent LLM, especially AI Scientist systems, show promising potential for automated ideation. However, existing approaches predominantly rely on pre-defined agentic workflows. This constraint severely limits the flexibility required to navigate the vast search space of scientific literature and the complex action space of research reasoning. Recently, training Agentic LLMs has emerged as a promising direction, offering flexible reasoning frameworks and the capability for autonomous tool utilization. However, there remains a non-trivial challenge: applying previous agentic data synthesis methods to scientific ideation suffers from prohibitively high data synthesis cost. To bridge this gap, we propose Agentic-Ideation, a novel framework comprising an automated trajectory synthesis pipeline and a specialized agentic LLM trained for scientific ideation. Specifically, we first define a comprehensive tool space incorporating three external tools and three cognitive tools. Then we introduce an Oracle-Guided Data Synthesis strategy. By leveraging a reference idea as oracle guidance, this approach steers the multi-agent system to efficiently reconstruct the logical reasoning and tool invocation paths, transforming aimless trial-and-error into directed trajectory generation. Finally, we train the agent on these synthesized trajectories, employing a masking strategy on tool execution results. This ensures the model focuses on decision-making logic without interference from external feedback. Experimental results demonstrate that our method outperforms the SOTA workflow-based baseline by \textbf{11.91\%} in overall quality. Furthermore, our approach improves the sample efficiency of high-quality data synthesis by \textbf{over 10$\times$}.

Problem

• scientific ideation
• agentic LLMs
• data synthesis cost
• sample efficiency
• reasoning flexibility

Innovation

• Agentic-Ideation
• Oracle-Guided Data Synthesis
• Sample Efficiency
• Scientific Ideation
• Tool-Augmented Reasoning