MOCA is a new modular transformer architecture for causal inference that applies one-way cross-attention and gradient detachment to keep treatment and outcome modeling separate, showing competitive or better performance than IPW, AIPW, X-learner, TARNet, and DragonNet on simulations and real health/
Mooij, David Sontag, Richard Zemel, and Max Welling
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ML 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
MOCA: A Transformer-based Modular Causal Inference Framework with One-way Cross-attention and Cutting Feedback
MOCA is a new modular transformer architecture for causal inference that applies one-way cross-attention and gradient detachment to keep treatment and outcome modeling separate, showing competitive or better performance than IPW, AIPW, X-learner, TARNet, and DragonNet on simulations and real health/