Top-K Structure Search with Solution Path
Manjunath, Sanjay
Manjunath, Sanjay
Author
Supervisor
Department
Machine Learning
Embargo End Date
2026-05-30
Type
Thesis
Date
2025
License
Language
English
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
Causal structure discovery algorithms often produce a single estimated causal graph without assessing its reliability. However, varying the sparsity regularization parameter (λ) in Lasso can lead to different graphical structures, making it crucial to explore how causal relationships evolve across different λ values. In this work, we propose a Top K Structure Search with Solution Path algorithm that systematically tracks the evolution of edge weights across a range of λ values. By ranking the resulting structures based on Likelihood or Bayesian Information Criterion (BIC) scores, our method identifies the Top K most plausible causal graphs. Unlike traditional approaches that return a single estimate, our method provides a ranked list of candidate structures, enabling better uncertainty assessment. We evaluate our approach on both synthetic and realworld datasets, demonstrating its effectiveness in capturing structural variations and improving causal discovery. Our results highlight the benefits of leveraging solution paths in causal inference, particularly in scenarios where a single estimated graph may not fully capture the underlying causal relationships. This work contributes to advancing structure learning by introducing a principled framework for ranking and interpreting causal graphs.
Citation
Sanjay Manjunath, “Top-K Structure Search with Solution Path,” Master of Science thesis, Machine Learning, MBZUAI, 2025.
Source
Conference
Keywords
Top-k, Causality, Causal Discovery, Uncertainty Estimation, Lasso, Solution Path