Innovation
Methods, ideas, or system contributions that make the work stand out.
Autoresearch Agents
Fork-and-Flush
Idea Basins
Parallel Trajectories
Context Reset
Fork-and-Flush: Escaping Idea Basins in Autoresearch Agents
🤖 AI Summary
This study addresses the performance stagnation of autonomous research agents on open-ended problems, where they frequently become trapped in local solution spaces. To overcome this limitation, we propose a "Branch and Refresh" strategy that introduces a novel periodic intervention mechanism. By combining workspace inheritance with context resetting, this approach guides parallel trajectory exploration to effectively escape cognitive basins and avoid local optima. Furthermore, it integrates embedding similarity analysis, parallel trajectory management, and dynamic evaluation selection to optimize long-horizon task performance. Experimental results across 13 long-horizon tasks demonstrate that, under equivalent computational budgets, the proposed method improves average scores by 66.0% over single-trajectory baselines and by 44.4% compared to Best-of-N approaches.
📝 Abstract
Autoresearch agents tackle open-ended problems by repeatedly proposing candidate solutions, evaluating them, and using feedback to guide subsequent experiments. We show that independent runs of the same agent on the same task often plateau at substantially different scores, with gaps that persist even after considerable additional compute. Embedding their candidate artifacts by functional similarity provides further evidence that trajectories remain in localized regions of the solution space, which we call idea basins. To help agents escape these basins, we study a simple periodic intervention, fork-and-flush. Our method forks the agent into parallel trajectories, each inheriting the accumulated workspace but starting with a fresh chat context. After running each trajectory for a fixed horizon, the agent continues from the highest-scoring one. Across 13 long-horizon research and engineering tasks, with individual agent runs lasting up to several days, fork-and-flush outperformed the single-run and best-of-N baselines by a relative improvement of 66.0% and 44.4%, respectively, on the min-max normalized average score under an equal compute budget.
🔎 Similar Papers
No similar papers found.
