Nonparametric Expert DAG Learning with Accurate Edge Strengths and Realistic Knowledge Incorporation
Keywords: probabilistic inference, nonparametric method, knowledge representation
Abstract: Directed Acyclic Graphs (DAGs) are crucial for modeling causal structures and complex dependencies in domains such as biology, healthcare, and finance. Effective structure learning must not only align with domain expert knowledge but also produce interpretable model decisions. Though continuous structure learning methods like NOTEARS are gaining popularity, an underexplored feature is their ability to open up the black box of decisions made by traditional combinatorial search by quantifying edge strengths in weighted adjacency matrices. Yet challenges persist in systematically integrating expert knowledge and ensuring learned weights accurately reflect true edge relationships. We present Non-parametric Expert DAG (NEDAG), a novel method that formulates accurate weight matrices using Gaussian Processes (GPs) and incorporates realistic domain knowledge into the continuous structure learning framework. Experiments on both synthetic and real-world datasets demonstrate that NEDAG not only surpasses existing methods in structure accuracy but also produces more accurate edge strengths. NEDAG thus provides a robust and interpretable solution for structure discovery in real-world applications.
Primary Area: learning on graphs and other geometries & topologies
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 12944
Loading