Statistics Group



Online inference

Anagnostopoulos, Cosma, Dawid, de Rooij, Gramacy, Singh, Yildirim

Advances in data collection technology have enabled acquisition of real-time information in various domains, including Internet traffic, high-frequency financial data, target tracking and the biological sciences. Online inference addresses the need to extract information efficiently from sequentially observed systems.

  • Probabilistic counting algorithms for streaming data.

  • Collaboration with Engineering in the area of Particle Filters, the gold-standard recursive computational procedure for the widely used Hidden Markov Models.

  • Novel online estimation technique for changepoint systems, to handle heterogeneity of sequential data.

  • Combining sequential predictions of experts for forecasting, and Bayesian Model Averaging.

  • Handling temporal variation in streaming classification via self-tuning learning rates.

  • Dynamic trees for learning and design.

  • Using particle filters for online inference and variable selection for high-dimensional Bayesian regularized regression.

Bobby2
Dynamic trees utilise the strength of particle filtering to fit tree-structured nonparametric regression models to sequential data. The plots show posterior mean and 90\% interval for each of 30 runs with 1000 particles.
rocstream
The evolution of the trade-off between false positives and false negatives (ROC curve analysis) for a streaming classifier with self-tuning forgetting.

  1. Clifford, P. and Cosma, I. A. (2011)
    A statistical analysis of probabilistic counting algorithms
    Scandinavian Journal of Statistics, to appear.

  2. Dawid, A.P., de Rooij, S., Shafer, G., Shen, A., Vereshchagin, N., and Vovk, V. (2011)
    Insuring against loss of evidence in game-theoretic probability.
    Statistics and Probability Letters}, to appear.

  3. Gramacy, B., Taddy, M. and Polson, N. (2011)
    Dynamic trees for learning and design
    Journal of the American Statistical Association, to appear.

  4. Poyiadjis, G., Doucet, A. and Singh, S.S. (2011)
    Sequential Monte Carlo computation of the score and observed information matrix in state-space models with application to parameter estimation
    Biometrika, to appear.

  5. van Erven, T. , Grünwald, P.D. and de Rooij, S.
    Catching up faster by switching sooner: A predictive approach to adaptive estimation with an application to the AIC-BIC Dilemma.
    Under consideration as a Royal Statistical Society read paper.