TopoCurate: Modeling Interaction Topology for Tool-Use Agent Training

Jinluan Yang, Yuxin Liu, Zhengyu Chen, Chengcheng Han, Yueqing Sun, Qi Gu, Hui Su, Xunliang Cai, Fei Wu, Kun Kuang · 2026

Read TopoCurate: Modeling Interaction Topology for Tool-Use Agent Training by Jinluan Yang, Yuxin Liu, Zhengyu Chen, Chengcheng Han, Yueqing Sun, Qi Gu, Hui Su, Xunliang Cai, Fei Wu, Kun Kuang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-01714.txt.

Description

Jinluan Yang, Yuxin Liu, Zhengyu Chen, and colleagues pick better training data for agents that use tools. Their 2026 work faults the usual filter, which keeps only winning runs and tasks with middling pass rates. A winning run may still waste steps or never show recovery from an error, so outcome alone misleads.

Their frame, TopoCurate, merges many trial runs of one task into a single map of shared states. On that map, good runs stand out by recovery skill, lean steps, and varied strategy. For practice tasks, the frame favors tasks rich in error branches and mixed plans. Tests on two agent benchmarks show steady gains for both fine-tuning and reward-based training. The map makes interaction structure, not just wins, drive what agents learn from.

Catalog record

CREATORJinluan Yang, Yuxin Liu, Zhengyu Chen, Chengcheng Han, Yueqing Sun, Qi Gu, Hui Su, Xunliang Cai, Fei Wu, Kun Kuang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE69168
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.01714

Browse the full library · About the Tower of Babel