For papers in chronological order, see here.

(* indicates equal contributions)

Model selection in non-standard settings

In problems like variable selection or graph estimation, models have simple Boolean structures—variables or edges being present or absent -- making errors easy to define and control. However, many modern models, such as those based on partitions, permutations, directed acyclic graphs, subspaces, or phylogenetic trees, lack this simple structure, complicating the formalization and control of model selection errors. We address this by representing model spaces as partially ordered sets, providing a unifying, principled framework for model selection.
• "Consensus Tree Estimation with False Discovery Rate Control via Partially Ordered Sets"
       M. Valdez, A. Willis, A. Taeb
       2025.
• "Model-oriented Graph Distances via Partially Ordered Sets"
       A. Taeb *, F.R. Guo*, L. Henckel*
       2025.
• "Quantifying Uncertainty and Stability Among Highly Correlated Predictors: a Subspace Perspective"
       X. Zhang, J. Bien, A. Taeb [software]
       2025.
• "Model Selection over Partially Ordered Sets"
       A. Taeb , P. Bühlmann, V. Chandrasekaran [software][poster] [video].
       Proceedings of National Academy of Sciences, 2024.
• "False Discovery and Its Control in Low-rank Estimation"
       A. Taeb , P. Shah, V. Chandrasekaran, [software] [video].
       Journal of the Royal Statistical Society (Series B), 2020.

Causal inference: discovery and robust predictions

Causal reasoning is often seen as the ultimate goal of science. My work in this area follows two threads. The first focuses on causal discovery — the challenge of reliably uncovering causal relationships from data. Existing optimization-based methods often learn suboptimal causal models. I address this by leveraging advances in convex mixed-integer programming . The second thread studies domain adaptation, the problem of learning prediction models that generalize out-of-distribution. By targeting underlying causal mechanisms, we develop methods “geared towards causality” that achieve robust predictions beyond the training data, while relying on weaker assumptions than full causal identifiability
• "Convex Mixed-Integer Programming for Causal Additive Models with Optimization and Statistical Guarantees "
       X. Zhang, N. Keret, A. Shojaie, A. Taeb [software]
       2025.
• "An Asymptotically Optimal Coordinate Descent Algorithm For Learning Bayesian Networks from Gaussian Models "
       T. Xu, S. Küçükyavuz, A. Shojaie, A. Taeb [software]
       Journal of Machine Learning Research, 2025.
• "Causality-oriented Robustness: Exploiting General Noise Interventions"
       X. Shen, P. Bühlmann, A. Taeb [software][poster]
       Journal of American Statistical Association, 2025.
• "Integer Programming for Learning Directed Acyclic Graphs from Non-identifiable Gaussian Models "
       T. Xu*, A. Taeb *, S. Küçükyavuz, A. Shojaie [software]
       Biometrika, 2025.
• "Characterization and Greedy Learning of Gaussian Structural Causal Models under Unknown Interventions "
       J. Gamella, A. Taeb , C. Heinze-Deml, P. Bühlmann, [software]
       2025.
• "A Fast Non-parametric Approach for Local Causal Structure Learning"
       M. Azadkia, A. Taeb , P. Bühlmann
       2022.

Latent-variable modeling: algorithms and applications

Many driving factors of physical systems are latent or unobserved. Thus, understanding such systems and producing robust predictions crucially relies on accounting for the influence of the latent structure. My work addresses methodological and theoretical advances in important problems in latent-variable modeling.
• "A Spectral Method for Multi-view Subspace Learning Using the Product of Projections "
       R. Sergazinov, A. Taeb , I. Gaynanova
       Biometrika, 2025.
• "Extremal Graphical Modeling with Latent Variables via Convex Optimization"
       S. Engelke, A. Taeb [software]
       Journal of Machine Learning Research, 2025.
• "Learning and scoring Gaussian latent causal models with unknown additive interventions"
       A. Taeb , J. Gamella, C. Heinze-Deml, P. Bühlmann, [software] [video]
       Journal of Machine Learning Research, 2024.
• "Provable Concept Learning for Interpretable Predictions using Variational Inference"
       A. Taeb , N. Ruggeri, C. Schnuck, F. Yang, [software][video][poster]
       ICML Workshop on AI4Science,2022.
• "Learning Exponential Family Graphical Models with Latent Variables using Regularized Conditional Likelihood"
       A. Taeb , P. Shah, V. Chandrasekaran
       2020.
• "Interpreting Latent Variables in Factor Models via Convex Optimization"
       A. Taeb , V. Chandrasekaran
       Mathematical Programming, 2018.
• "A Statistical Graphical Model of the California Reservoir Network"
       A. Taeb , J.T. Reager, M. Turmon, V. Chandrasekaran [software] [press release] [broad audience]
       Water Resources Research, 2017.
• "Visual Stylometry using background selection and wavelet-HMT-based Fisher Information distances"
       H. Qi, A. Taeb , and S. Hughes
       EURASIP Signal Processing, 2012.