Publications
Preprints:
Semiparametric Triple Difference Estimators
TL;DR. Triple differences can relax parallel trends by borrowing information from an auxiliary domain. We give identification formulas and doubly robust semiparametric estimators for both panel and repeated cross-section data. The repeated cross-section setup also allows covariates to change over time.
@misc{akbari2025semiparametric,
title = {Semiparametric Triple Difference Estimators},
author = {Sina Akbari and Negar Kiyavash and AmirEmad Ghassami},
year = {2025},
eprint = {2502.19788},
archivePrefix= {arXiv},
primaryClass = {econ.EM},
url = {https://arxiv.org/abs/2502.19788}
}
Journal publications:
JMLR 2025 Optimal Experiment Design for Causal Effect Identification Journal of Machine Learning Research (JMLR) special issue for ICML&NeurIPS outstanding papers.
TL;DR. When observations alone cannot identify a causal effect, which interventions should we pay for? We cast minimum-cost identification as a hitting-set problem, prove the exact problem is NP-complete, and give optimal, approximate, and fast heuristic methods. In simulations, the heuristics stay close to the minimum cost.
@article{akbari2025optimal,
author = {Sina Akbari and Jalal Etesami and Negar Kiyavash},
title = {Optimal Experiment Design for Causal Effect Identification},
journal = {Journal of Machine Learning Research},
year = {2025},
volume = {26},
number = {28},
pages = {1--56},
url = {https://www.jmlr.org/papers/v26/22-1516.html}
}
JMLR 2025 Recursive Causal Discovery
TL;DR. Causal discovery gets brittle and expensive when conditional-independence tests use large conditioning sets. The removable-variable view shrinks the graph one variable at a time, cutting both the number and size of those tests. The framework nearly matches worst-case lower bounds and is implemented in the RCD Python package.
@article{JMLR:v26:24-0384,
author = {Ehsan Mokhtarian and Sepehr Elahi and Sina Akbari and Negar Kiyavash},
title = {Recursive Causal Discovery},
journal = {Journal of Machine Learning Research},
year = {2025},
volume = {26},
number = {61},
pages = {1--65},
url = {https://www.jmlr.org/papers/v26/24-0384.html}
}
TL;DR. Influence-function corrections can improve flexible machine-learning estimates of the average treatment effect, but there is no universal free lunch. Across datasets, the gains depend on the estimator, sample size, and data-generating process. The practical message is to stress-test conclusions across multiple estimation pipelines.
@article{vowels2023free,
title = {A Free Lunch with Influence Functions? An Empirical Evaluation of Influence Functions for Average Treatment Effect Estimation},
author = {Matthew J. Vowels and Sina Akbari and Necati Cihan Camgoz and Richard Bowden},
journal = {Transactions on Machine Learning Research},
year = {2023},
url = {https://openreview.net/forum?id=dQxBRqCjLr}
}
JMLR 2021 A Recursive Markov Boundary-Based Approach to Causal Structure Learning
TL;DR. Use local Markov-boundary information to find a variable that can be safely peeled off, learn its neighborhood, and recurse. This sharply reduces the conditional-independence testing burden of constraint-based causal discovery. The resulting complexity is close to a lower bound and performs well on synthetic and real graphs.
@InProceedings{pmlr-v150-mokhtarian21a,
title = {A Recursive Markov Boundary-Based Approach to Causal Structure Learning},
author = {Mokhtarian, Ehsan and Akbari, Sina and Ghassami, AmirEmad and Kiyavash, Negar},
booktitle = {Proceedings of The KDD'21 Workshop on Causal Discovery},
pages = {26--54},
year = {2021},
volume = {150},
series = {Proceedings of Machine Learning Research},
publisher = {PMLR},
url = {https://proceedings.mlr.press/v150/mokhtarian21a.html}
}
Conference publications:
UAI 2026 ORAL Best Paper Award Fundamental Limits and Optimal Methods for Sharp Analytical Causal Bounds in Instrumental Variable Models
TL;DR. Sharp analytical instrumental-variable bounds cannot stay simple as the outcome alphabet grows. We prove that any sharp bound needs exponentially many linear pieces, and the same exponential barrier appears in the IV inequalities. We also give Python and R procedures that attain this optimal complexity.
@InProceedings{pmlr-v337-boushehrian26a,
title = {Fundamental Limits and Optimal Methods for Sharp Analytical Causal Bounds in Instrumental Variable Models},
author = {Boushehrian, Arefe and Badri, Mohammad Reza and Akbari, Sina and Kiyavash, Negar},
booktitle = {Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence},
pages = {689--744},
year = {2026},
volume = {337},
series = {Proceedings of Machine Learning Research},
publisher = {PMLR},
url = {https://proceedings.mlr.press/v337/boushehrian26a.html}
}
TL;DR. Standard neural networks can ignore causal structure that we already know. CaTs and causal fully connected networks bake a user-specified DAG into the architecture, so predictions respect the allowed causal pathways. The aim is better robustness under distribution shift and clearer interpretation without giving up expressive models.
@inproceedings{vowels2026cats,
title = {CaTs and DAGs: Integrating Directed Acyclic Graphs with Transformers for Causally Constrained Predictions},
author = {Matthew James Vowels and Mathieu Rochat and Sina Akbari},
booktitle = {The Fourteenth International Conference on Learning Representations},
year = {2026},
url = {https://openreview.net/forum?id=ZIQactmQxb}
}
UAI 2025 Causal Effect Identification in Heterogeneous Environments from Higher-Order Moments
TL;DR. Multiple environments can reveal a causal effect even with latent confounding, if the effect stays invariant and the heterogeneity is structured. Higher-order moments identify and estimate the effect when only one part of the data-generating mechanism changes across environments. If both the latent and treatment noise distributions change, identification can fail.
@InProceedings{pmlr-v286-kivva25a,
title = {Causal Effect Identification in Heterogeneous Environments from Higher-Order Moments},
author = {Kivva, Yaroslav and Akbari, Sina and Salehkaleybar, Saber and Kiyavash, Negar},
booktitle = {Proceedings of the Forty-first Conference on Uncertainty in Artificial Intelligence},
pages = {2223--2254},
year = {2025},
volume = {286},
series = {Proceedings of Machine Learning Research},
publisher = {PMLR},
url = {https://proceedings.mlr.press/v286/kivva25a.html}
}
UAI 2025 Multi-armed Bandits with Missing Outcomes
TL;DR. Missing rewards are not harmless in bandits: ignoring why outcomes disappear can drive regret to linear. We develop algorithms for both missing-at-random and missing-not-at-random feedback and show how the missingness mechanism changes achievable regret. Accounting for it gives much better decision-making guarantees.
@InProceedings{pmlr-v286-mahrooghi25a,
title = {Multi-armed Bandits with Missing Outcomes},
author = {Mahrooghi, Ilia and Moradi, Mahshad and Akbari, Sina and Kiyavash, Negar},
booktitle = {Proceedings of the Forty-first Conference on Uncertainty in Artificial Intelligence},
pages = {2844--2875},
year = {2025},
volume = {286},
series = {Proceedings of Machine Learning Research},
publisher = {PMLR},
url = {https://proceedings.mlr.press/v286/mahrooghi25a.html}
}
TL;DR. Can we tell two continuous distributions apart without parametric assumptions, and with how many samples? A von Mises KL estimator gives an optimal-rate closeness test under smoothness assumptions. Plugging that test into causal discovery yields finite-sample guarantees even for nonlinear, multidimensional variables with hidden confounding.
@InProceedings{pmlr-v275-jamshidi25a,
title = {Sample Complexity of Nonparametric Closeness Testing for Continuous Distributions and Its Application to Causal Discovery with Hidden Confounding},
author = {Jamshidi, Fateme and Akbari, Sina and Kiyavash, Negar},
booktitle = {Proceedings of the Fourth Conference on Causal Learning and Reasoning},
pages = {1217--1238},
year = {2025},
volume = {275},
series = {Proceedings of Machine Learning Research},
publisher = {PMLR},
url = {https://proceedings.mlr.press/v275/jamshidi25a.html}
}
NeurIPS 2024 Fast Proxy Experiment Design for Causal Effect Identification
TL;DR. Direct interventions may be too expensive, so intervene on cheaper proxy variables instead. We reformulate the minimum-cost proxy experiment problem to make exact optimization much faster in practice, and also study designs based on valid adjustment sets. The same identification goal can be reached with far less computational overhead.
@inproceedings{elahi2024fast,
title = {Fast Proxy Experiment Design for Causal Effect Identification},
author = {Sepehr Elahi and Sina Akbari and Jalal Etesami and Negar Kiyavash and Patrick Thiran},
booktitle = {Advances in Neural Information Processing Systems},
volume = {37},
pages = {51254--51283},
year = {2024},
url = {https://proceedings.neurips.cc/paper_files/paper/2024/hash/5bd9fbb3a5a985f80c16ddd0ec1dfc43-Abstract.html}
}
ICML 2024 SPOTLIGHT Triple changes estimator for targeted policies
TL;DR. Triple differences relax parallel trends, while changes-in-changes recovers an entire counterfactual distribution rather than just a mean. We combine the two into a triple-changes estimator and extend it to multivariate outcomes using optimal transport. The method is illustrated on Medicaid expansion and children's preventive care.
@InProceedings{pmlr-v235-akbari24a,
title = {Triple Changes Estimator for Targeted Policies},
author = {Akbari, Sina and Kiyavash, Negar},
booktitle = {Proceedings of the 41st International Conference on Machine Learning},
pages = {666--695},
year = {2024},
volume = {235},
series = {Proceedings of Machine Learning Research},
publisher = {PMLR},
url = {https://proceedings.mlr.press/v235/akbari24a.html}
}
NeurIPs 2023 Causal effect identification in uncertain causal networks
TL;DR. Causal identification usually assumes the graph is known exactly. Here each edge can be uncertain, and the task is to find the most plausible subgraph in which the target effect is identifiable. That optimization is NP-hard, so we develop efficient approximations and test them on synthetic and real networks.
@inproceedings{akbari2023uncertain,
title = {Causal Effect Identification in Uncertain Causal Networks},
author = {Sina Akbari and Fateme Jamshidi and Ehsan Mokhtarian and Matthew Vowels and Jalal Etesami and Negar Kiyavash},
booktitle = {Advances in Neural Information Processing Systems},
volume = {36},
year = {2023},
url = {https://proceedings.neurips.cc/paper_files/paper/2023/hash/017c897b4d85a744f345ccbf9d71e501-Abstract-Conference.html}
}
NeurIPS 2023 Causal imitability under context-specific independence relations
TL;DR. Imitation can fail when hidden causal structure makes the expert's behavior impossible to reproduce from observed data. Known context-specific independences can change that answer. We characterize when they help, show the general feasibility problem is NP-hard, and give a sound algorithm that uses both CSI information and data.
@inproceedings{jamshidi2023imitability,
title = {Causal Imitability Under Context-Specific Independence Relations},
author = {Fateme Jamshidi and Sina Akbari and Negar Kiyavash},
booktitle = {Advances in Neural Information Processing Systems},
volume = {36},
year = {2023},
url = {https://proceedings.neurips.cc/paper_files/paper/2023/hash/54e13b23fa2f399cea6e67acf9063c40-Abstract-Conference.html}
}
ICML 2022 ORAL Outstanding Paper Runner-up Award Minimm-cost Intervention Design for Causal Effect Identification.
TL;DR. When observational data cannot identify the effect you want, the next question is: what is the cheapest experiment that will? We connect minimum-cost intervention design to the minimum hitting-set problem and give exact, approximation, and polynomial-time heuristic algorithms. The heuristics achieve low regret on random causal graphs.
@InProceedings{pmlr-v162-akbari22a,
title = {Minimum Cost Intervention Design for Causal Effect Identification},
author = {Akbari, Sina and Etesami, Jalal and Kiyavash, Negar},
booktitle = {Proceedings of the 39th International Conference on Machine Learning},
pages = {258--289},
year = {2022},
volume = {162},
series = {Proceedings of Machine Learning Research},
publisher = {PMLR},
url = {https://proceedings.mlr.press/v162/akbari22a.html}
}
AAAI 2022 Learning Bayesian Networks in the Presence of Structural Side Information
TL;DR. Structural side information can make Bayesian-network learning much cheaper. We build recursive constraint-based algorithms for bounded clique number and diamond-free graphs, showing that bounded-treewidth networks can be learned with polynomially many conditional-independence tests. Experiments show strong gains in scalability.
@inproceedings{mokhtarian2022learning,
title = {Learning Bayesian Networks in the Presence of Structural Side Information},
author = {Ehsan Mokhtarian and Sina Akbari and Fateme Jamshidi and Jalal Etesami and Negar Kiyavash},
booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence},
volume = {36},
number = {7},
pages = {7814--7822},
year = {2022},
doi = {10.1609/aaai.v36i7.20750},
url = {https://ojs.aaai.org/index.php/AAAI/article/view/20750}
}
NeurIPS 2021 Recursive Causal Structure Learning in the Presence of Latent Variables and Selection Bias.
TL;DR. Latent variables and selection bias make constraint-based causal discovery especially expensive. We identify removable variables in a MAG, peel them off recursively, and still retain soundness and completeness. This reduces both the number of conditional-independence tests and the size of their conditioning sets, with near-tight worst-case complexity.
@inproceedings{akbari2021recursive,
title = {Recursive Causal Structure Learning in the Presence of Latent Variables and Selection Bias},
author = {Sina Akbari and Ehsan Mokhtarian and AmirEmad Ghassami and Negar Kiyavash},
booktitle = {Advances in Neural Information Processing Systems},
volume = {34},
pages = {10119--10130},
year = {2021},
url = {https://proceedings.neurips.cc/paper/2021/hash/53edebc543333dfbf7c5933af792c9c4-Abstract.html}
}
Workshop Publications:
NeurIPS 2023 — workshop on Optimal Transport and Machine Learning (OTML) Causal Discovery via Monotone Triangular Transport Maps
TL;DR. Monotone triangular transport maps turn distribution fitting into a tool for causal discovery. They give noise-agnostic conditional-independence tests for constraint-based learning and a new score for score-based learning. With stronger model assumptions, the same framework can identify a unique causal graph.
@inproceedings{akbari2023transport,
title = {Causal Discovery via Monotone Triangular Transport Maps},
author = {Sina Akbari and Luca Ganassali and Negar Kiyavash},
booktitle = {NeurIPS 2023 Workshop on Optimal Transport and Machine Learning},
year = {2023},
url = {https://openreview.net/forum?id=PMGGrKTIii}
}