TriOpt: A Scalable Algorithm for Linear Causal Discovery
Published in arXiv, 2026
Recommended citation: @article{Joy2026TriOpt, author = {Joy, Rafat Ashraf and Zheleva, Elena}, title = {{TriOpt: A Scalable Algorithm for Linear Causal Discovery}}, journal = {arXiv preprint arXiv:2605.17465}, year = {2026}, month = {May}, doi = {10.48550/arXiv.2605.17465} } [https://arxiv.org/abs/2605.17465](https://arxiv.org/abs/2605.17465)
Learning causal relations from observational data is challenging because the graph search space grows super-exponentially with the number of variables. TriOpt integrates ordering-based and continuous optimization approaches into two efficient stages: fast topological ordering estimation using the Sherman-Morrison rank-1 downdate and additive linear kernels, followed by convex structure learning without an acyclicity constraint. Under the true ordering, TriOpt exactly recovers the underlying linear DAG, while experiments on synthetic, semi-synthetic, and real-world datasets demonstrate orders-of-magnitude speedups with comparable or superior accuracy.
Recommended citation: @article{joy2026triopt, title={TriOpt: A Scalable Algorithm for Linear Causal Discovery}, author={Joy, Rafat Ashraf and Zheleva, Elena}, journal={arXiv preprint arXiv:2605.17465}, year={2026} }