On the Approach of Ensemble of Interacting Imperfect Models
Several approaches of ensemble of interacting imperfect models combined based on observed data either by adaptive synchronization, optimized couplings or weighted combining have been recently proposed. In this study we further examine the weighted combining method using the Hindmarsh-Rose (HR) neuron model and the different outcomes that we can expect. We generate data with an HR model usually referred as ‘truth’ and use the data to train an ensemble of HR models with perturbed parameter values, so that together they mimic the truth. The results show that the weights of the ensemble can be learned using data from a truth HR model exhibiting bursting, in order to represent the same bursting behavior as well as other behaviors such as spiking and random bursting.
KeywordsConvex Combination Coupling Coefficient Neuron Model Truth Behavior Imperfect Model
This work was partially supported by project EC Grant #266722.
- 1.C.M. Bishop, Pattern Recognition and Machine Learning (Information Science and Statistics) (Springer, Secaucus, 2006)Google Scholar
- 2.G. Duane, J. Tribbia, B. Kirtman, Consensus on long-range prediction by adaptive synchronization of models, in EGU General Assembly Conference Abstracts, ed. by D.N. Arabelos, C.C. Tscherning, pp. 13324, April 2009Google Scholar
- 3.G. Duane, Synchronicity from synchronized chaos. Arxiv.org/abs/1101.2213. (Submitted 2011)Google Scholar
- 5.M. Mirchev, G.S. Duane, W.K.S. Tang, L. Kocarev, Improved modeling by coupling imperfect models. Commun. Nonlinear Sci. Numer. Simul. 17(7), 2741–2751 (2012)Google Scholar
- 6.W. Wiegerinck, F.M. Selten, Supermodeling: combining imperfect models through learning, in NIPS Workshop on Machine Learning for Sustainability (MLSUST), 2011Google Scholar