The authors generated anonymous datasets from six causal mechanism families, ranging from linear Gaussian relationships to difficult heavy-tailed settings.
LLM agents can repair causal graphs beyond algorithm selection
Heuresis isolates data-driven graph repair by removing variable names, preserving algorithmic headroom and tracing every restored edge.
Big Tech
Pingchuan Ma · Rui Ding · Shuai Wang
Zhejiang University of Technology · Microsoft · The Hong Kong University of Science and Technology
Research Digest··3 min read
Ma, Ding and Wang introduce Heuresis, a benchmark designed to test whether LLM agents can improve causal graphs through targeted analysis rather than prior knowledge or simple algorithm selection.
§