Causal Generative Neural Networks

Abstract

We present Causal Generative Neural Networks (CGNNs) to learn functional causal models from observational data. CGNNs leverage conditional independencies and distributional asymmetries to discover bivariate and multivariate causal structures. CGNNs make no assumption regarding the lack of confounders, and learn a differentiable generative model of the data by using backpropagation. Extensive experiments show their good performances comparatively to the state of the art in observational causal discovery on both simulated and real data, with respect to cause-effect inference, v-structure identification, and multivariate causal discovery.

References in corpus (5)

• Adam: A Method for Stochastic Optimization
• Generative Moment Matching Networks
• Kernel-based Conditional Independence Test and Application in Causal Discovery
• Distinguishing cause from effect using observational data: methods and benchmarks
• From Dependence to Causation

Cited by in corpus (12)

• Causality for Machine Learning
• Causal Discovery in Physical Systems from Videos
• Learning Neural Causal Models from Unknown Interventions
• Domain Adaptation as a Problem of Inference on Graphical Models
• Size-Invariant Graph Representations for Graph Classification Extrapolations
• FASK with Interventional Knowledge Recovers Edges from the Sachs Model
• Distinguishing Cause from Effect Using Quantiles: Bivariate Quantile Causal Discovery
• Amortized learning of neural causal representations
• Efficient Neural Causal Discovery without Acyclicity Constraints
• Neural Networks for Learning Counterfactual G-Invariances from Single Environments
• A Meta Learning Approach to Discerning Causal Graph Structure
• Adversarial Orthogonal Regression: Two non-Linear Regressions for Causal Inference