RefCon improves memory extraction for evolving AI agents
RefCon: Iterative Refinement and Contrastive Memory Extraction for Context-Evolving Agent
Artificial IntelligenceMachine LearningMultiagent Systems
Summary
When AI agents interact over long periods, they collect lots of information that can be noisy and hard to use. The authors propose RefCon, a method that helps these agents improve their memory by refining and comparing what they remember without needing perfect answers as guidance. This method works better than previous ones on different tasks, including software engineering, and keeps getting better with more data. It also balances accuracy and efficiency well.
What this means in practice
- •For software engineers: Enhance AI tools that assist coding by improving memory use without relying on labeled training data.
- •For ai development teams: Improve long-term AI agent performance by iteratively refining memory from ongoing interactions.
Authors
Ubaidillah Ariq Prathama, Bo Liu, Yeo Boon Hong, Yu-Xuan Huang, Yangkai Ding, Tao Yu
Abstract
Long-horizon agent interactions generate useful but noisy experience, and retraining models to absorb it is expensive. Context-evolving agents therefore need memory extraction methods that improve with more test-time compute without relying on gold labels. We propose RefCon, which combines sequential self-refinement with parallel self-contrast to extract higher-quality memories without gold labels. Evaluated on AppWorld and BFCL-V3 across multiple context-evolving agent frameworks, RefCon delivers strong and consistent gains, including relative improvements of 21.6% on ACE and 16.6% on ReMe over no-scaling baselines, while a diversity-focused variant (DivCon) achieves a 35.5% gain on ReasoningBank. RefCon consistently outperforms existing baselines without ground-truth labels, and generalizes across model scales and to software engineering tasks, where it surpasses even ground-truth baselines. We further analyze the accuracy-token trade-off and scaling behavior, showing RefCon maintains favorable efficiency and continues to improve as more trajectories are used, unlike diversity-only scaling which saturates earlier.