ADEPT is an EPSRC sponsored project between the
Universities of
Manchester and Bristol.
ADEPT (Adaptive Dynamic Ensemble Prediction
Techniques) aims to capitalize on the synergistic interface between three fields:
evolutionary computation, ensemble learning, and probabilistic modeling.
The project's primary goal is to take a heuristic technique from
the Evolutionary Computation literature - Learning Classifier
Systems - and translate it into an ensemble-based probabilistic model.
The probabilistic model we have devloped can precisely reproduce the capabilities of the LCS - an online
supervised learning
system, continuously adaptive, maintaining a parsimonious set of human-interpretable rules. However, the new
model stands apart from
the parameter-laden heuristic nature of LCS, having the advantages of a statistical underpinning:
flexibility and a solid
probabilistic foundation.
In the course of understanding how to construct this model, we have also contributed significantly to
the understanding of ensemble diversity in nonstationary learning, and information theoretic feature
selection
methodologies.
Principal Investigators:
Gavin Brown (Manchester),
Tim Kovacs (Bristol)
Research Staff: Narayanan Edakunni,
Ming-Jie Zhao
PhDs: Richard
Stapenhurst
News::
Richard's paper was accepted to IEEE CIDUE, and Gavin
will be delivering a Keynote address at the symposium.
Publications:
Online, GA based Mixture of Experts : a Probabilistic Model of UCS
Nara Edakunni, Gavin Brown, Tim Kovacs
Proceedings of the Genetic and Evolutionary Computation COnference (GECCO). July 2011
Analysis of Accuracy Discounting in UCS and its Effect on Voting Margins
Tim Kovacs, Nara Edakunni, Gavin Brown
Proceedings of the Genetic and Evolutionary Computation COnference (GECCO). July 2011
From Heuristics to Statistics: An Overview of the ADEPT project
Keynote address at IEEE Symposium on Computational Intelligence in Dynamic and Uncertain Environments.
Paris, France, April
2011
Theoretical and Empirical Analysis of Diversity in Non-Stationary Learning
Richard Stapenhurst and Gavin Brown
IEEE Symposium on Computational Intelligence in Dynamic and Uncertain Environments. Paris, France, April
2011
Some thoughts at the interface of Ensemble Methods and Feature Selection
Gavin Brown
Invited talk at Intl Workshop on Multiple Classifier Systems. Cairo, April 2010
Modeling UCS as a Mixture of Experts
Nara Edakunni, Tim Kovacs, Gavin Brown, James
Marshall, Arjun Chandra
Proceedings of the Genetic and Evolutionary Computation COnference (GECCO). Montreal, Canada, July
2009
A New Perspective for Information Theoretic Feature Selection
Gavin Brown
Twelfth International Conference on Artificial Intelligence and Statistics. Florida, June 2009
UCSpv: Principled Voting in UCS Rule Populations
Gavin Brown, Tim Kovacs, James Marshall
Proceedings of the Genetic and Evolutionary Computation COnference (GECCO). July 2008
|
