Mike West


Arts and Sciences Distinguished Professor of Statistics and Decision Sciences

Some of Mike’s current and near-term research interests and projects relate to the following topics.

  1. Foundations:
    • Bayesian decision analysis and decision theory: New classes of utility functions in long-term forecasting and decision problems.
    • Foundational approaches to model uncertainty, calibration, comparison and combination, including the framework of Bayesian predictive synthesis (BPS) and approaches defined via goalspecific, decision theory-based model scoring.
    • Model and analysis scalability based on the “decouple/recouple” concept.
    • Approaches to defining and evaluating “variable prioritization/importance” in ranges of inference and decision problems in large-scale predictive models and network studies.
  2. Modelling:
    • Multivariate time series and dynamic modelling for analysis and forecasting of increasingly large-scale, structured data—including traditional time series and complex dynamic network data.
    • Scalable time series modelling using simultaneous dynamic graphical models and related approaches.
    • Scalable dynamic network modelling using decouple/recoupled dynamic network flow models.
    • Scalable models for multivariate discrete (count) time series.
    • Scalable, structured dynamic latent threshold models for complex, time-varying, multivariate, non-linear systems.
  3. Methodology and computation:
    • Computational scalability of dynamic models under topics 1 & 2 above. Approaches using new ideas for direct analytic and simulation methods enabled by new thinking about model formulation (and avoiding MCMC). Development in broad classes of non-linear dynamic models and in models of dynamic networks.
    • Methods of Bayesian emulation to enable efficient and scalable sequential Bayesian computation in dynamic models based on use of analytically tractable emulators.
    • Methods of Bayesian emulation for challenging problems of optimisation, to map “optimise” to “analyse” in synthetic Bayesian inference problems, allowing use of Bayesian simulation methods in broad classes of (dynamic, sequential, and other) optimisation problems.
    • GPU-based and other computational approaches able to capitalize on the decouple/recouple modelling strategies.
  4. Applications:
    • Macro-economic policy-related time series modelling, forecasting and decision analysis. Several collaborators in national central banking systems in USA, EU and Japan.
    • Financial time series modelling, forecasting and decision analysis (personal and corporate). Several collaborators in industry.
    • Large-scale forecasting and decision analysis in commercial contexts, including very high-dimensional problems in consumer industries, with collaborators in national and international companies. Growth areas include product demand, sales, hierarchical revenue systems.
    • Large-scale dynamic network data analysis, network characterization and anomaly detections. Collaborators in IT/ecommerce industries.