Learning General Causal Structures with a Hidden Dynamic Process for Climate Analysis
Fu, Minghao
Fu, Minghao
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
The study of learning causal structure with latent variables has advanced the understanding of the world by uncovering causal relationships and latent factors, e.g., Causal Representation Learning (CRL). However, in real-world scenarios, such as those in climate systems, causal relationships are often nonparametric, dynamic, and exist among both observed variables and latent variables. These challenges motivate us to consider a general setting in which causal relations are nonparametric and unrestricted in their occurrence, which is unconventional to current methods. To solve this problem, with the aid of 3-measurement in temporal structure, under mild assumptions, we theoretically show that both latent variables and processes can be identified from causally-related observations, subject to minor indeterminacy. Moreover, we tackle the general nonlinear Causal Discovery (CD) on observations, e.g., temperature, as a specific task of learning independent representation, with the principle of functional equivalence giving rise to this approach. Based on these insights, we develop an estimation approach simultaneously recovering both the observed causal structure and latent causal process. Simulation studies validate the theoretical foundations and demonstrate the effectiveness of the proposed methodology. In the experiments involving climate data, this approach offers a powerful and indepth understanding of the climate system.
Citation
Minghao Fu, “Learning General Causal Structures with a Hidden Dynamic Process for Climate Analysis,” Master of Science thesis, Machine Learning, MBZUAI, 2025.
Source
Conference
Keywords
Climate Analysis, Causal Discovery, Causal Representation Learning