Reproduce “Charts as Epistemic Actions for Experience-Driven Agentic Data Analysis” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xinzhe Wang, Xudong Weng, Zang Li, Shaolei Zhang
CHARTMIND teaches data agents when to use text, charts, or both by distilling reusable guidance from judged analytical trajectories, without updating model parameters. The benchmark shows that text and charts help on different task types, while free choice does not reliably select the stronger evidence. On DAComp-DA, GPT-5.6 Luna’s Overall score rises from 54.32 to 59.78, with higher token use.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.