Advances in Causal Discovery and Graph Learning

The field of causal discovery and graph learning is moving towards addressing complex challenges such as handling latent confounders, cyclic causal relationships, and imbalanced regression tasks. Researchers are developing innovative methods, including variants of existing algorithms and novel frameworks, to improve the accuracy and robustness of causal discovery and graph learning models. These advancements have the potential to enable more reliable and generalizable predictions, and to support applications in various domains, such as education, sociology, and policy modeling. Noteworthy papers in this area include Less Greedy Equivalence Search, which modifies the greedy step to reduce computational cost and improve finite-sample accuracy, and Bivariate Denoising Diffusion, which introduces a novel independence test statistic to handle latent noise introduced by unmeasured mediators. Other notable contributions include the development of Spectral Manifold Harmonization for addressing imbalanced regression challenges, and Relational Causal Discovery with Latent Confounders, which proposes a sound and complete causal discovery algorithm for relational data with latent confounders.

Sources

Less Greedy Equivalence Search

Prediction Gaps as Pathways to Explanation: Rethinking Educational Outcomes through Differences in Model Performance

When Additive Noise Meets Unobserved Mediators: Bivariate Denoising Diffusion for Causal Discovery

Strategic Counterfactual Modeling of Deep-Target Airstrike Systems via Intervention-Aware Spatio-Causal Graph Networks

$\sigma$-Maximal Ancestral Graphs

A Recipe for Causal Graph Regression: Confounding Effects Revisited

Spectral Manifold Harmonization for Graph Imbalanced Regression

Relational Causal Discovery with Latent Confounders

Built with on top of