580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. How can Bayesian network structure and parameter learning leverage local and global constraints for efficient and accurate causal model discovery?
This theme focuses on methods for learning Bayesian network structures and parameters that capture causal relationships by utilizing both local constraints (such as conditional probability distribution representations with decision trees or graphs) and global independence constraints. It also emphasizes Bayesian scoring functions and search algorithms that navigate complex search spaces, improving the accuracy and computational efficiency of causal Bayesian network discovery from data.
Key finding: This paper introduces a fully Bayesian scoring method for Bayesian networks that incorporate decision-graph representations of conditional probability distributions, extending beyond typical decision tree approaches. It... Read more
Key finding: The work presents comprehensive techniques for learning both the structure and parameters of Bayesian networks from data, combining prior expert knowledge and statistical learning. It highlights how Bayesian networks... Read more
Key finding: The paper proposes an elicitation and modeling approach that integrates expert variables into Bayesian networks without data, ensuring that the observed data distributions are preserved even when expert variables remain... Read more
2. What frameworks and computational techniques support causal inference and discovery in complex, high-dimensional, and heterogeneous data using causal Bayesian networks?
This research area investigates formal frameworks such as graphical causal modeling and structural equation models, combined with advanced computational techniques including causal machine learning, to infer causal relationships in settings characterized by large numbers of variables, complex temporal dynamics, hidden confounders, and hybrid data types. Emphasis is placed on bridging theoretical causal inference with practical algorithmic realizations, addressing challenges of generalization, data heterogeneity, and combined observational and experimental data.
Key finding: This foundational overview introduces graphical causal modeling as a principled framework for causal discovery from observational and experimental data, emphasizing the development of well-defined causal structures,... Read more
Key finding: The paper shows how causal machine learning (CML) integrates graphical causal models with advanced machine learning methods to handle high-dimensional, multimodal, and temporal healthcare data. It categorizes CML into causal... Read more
Key finding: This study introduces a practical causal discovery algorithm using causal Bayesian networks tailored for biomedical Big Data that integrates clinical, genomics, proteomics, and environmental datasets. It addresses challenges... Read more
Key finding: The paper proposes hybrid semiparametric Bayesian networks that combine parametric models (conditional linear Gaussian assumptions) and nonparametric estimation to capture complex dependencies in hybrid discrete-continuous... Read more
3. How do cognitive and philosophical analyses enhance the understanding and validity of causal Bayesian network applications in reasoning and intervention?
This theme explores the conceptual foundations of causation as represented in causal Bayesian networks, linking computational models with human causal cognition and philosophical arguments. It includes investigations into the nature of token versus type-level causation, causal exclusion and efficacy arguments framed in Bayesian networks, and how sampling-based cognitive models approximate causal judgments. These insights inform the interpretation, justification, and refinement of causal Bayesian networks for both AI modeling and explanatory purposes.
Key finding: This paper reconstructs causal exclusion arguments within the formal framework of causal Bayesian networks, showing that supervenience relations behave analogously to causal relations under the causal Markov and minimality... Read more
Key finding: Extending the Mutation Sampler cognitive model, this work incorporates prior distributions into Bayesian sampling to better explain the variability and skewness observed in human probabilistic causal judgments relative to... Read more
Key finding: The paper analyzes the relation between type-level causation encoded by causal Bayesian networks and token-level actual causation theories. It argues that Bayesian networks provide robust, abstract representations of causal... Read more
Key finding: This experimental study finds that people often engage in transitive causal reasoning consistent with the Markov condition when interpreting causal chains, even when presented with data that violate this condition.... Read more