GO-CBED trains a transformer policy to choose intervention sequences that maximize expected information gain on a user-specified causal query, using a variational bound with normalizing-flow posteriors, and reports gains over baselines in causal reasoning and discovery benchmarks.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Goal-Oriented Sequential Bayesian Experimental Design for Causal Learning
GO-CBED trains a transformer policy to choose intervention sequences that maximize expected information gain on a user-specified causal query, using a variational bound with normalizing-flow posteriors, and reports gains over baselines in causal reasoning and discovery benchmarks.