Biomedical subjects

P M Granitto

Publications and source records attributed to P M Granitto.

3 recordsLinked to original sources

PubMed2006

Overembedding method for modeling nonstationary systems.

We propose a general overembedding method for modeling and prediction of nonstationary systems. It basically enlarges the standard time-delay-embedding space by inclusion of the (unknown) slow driving signal, which is estimated simultaneously with the intrinsic stationary dynamics. Our method can be implemented with any modeling tool. Using, in particular, artificial neural networks, its application to both synthetic and real-world time series shows that it is highly efficient, leading to much more accurate results and longer prediction horizons than other existing overembedding methods in the literature.

Algorithms↗

PubMed2001

Nonstationary time-series analysis: accurate reconstruction of driving forces.

We propose a simple method for the accurate reconstruction of slowly changing external forces acting on nonlinear dynamical systems. The method traces the evolution of the external force by locally linearizing the map dependency with the shifting parameter. Application of our algorithm to synthetic data corresponding to discrete models of evolving ecosystems shows an accuracy that outperforms those of previous methods in the literature. In addition, an application to the real-world sunspot time series recovers recently reported changes in solar activity during the last century.

Journal Article↗

PubMed2001

A late-stopping method for optimal aggregation of neural networks.

Ensembles of artificial neural networks have been used in the last years as classification/regression machines, showing improved generalization capabilities that outperform those of single networks. However, it has been recognized that for aggregation to be effective the individual networks must be as accurate and diverse as possible. An important problem is, then, how to tune the aggregate members in order to have an optimal compromise between these two conflicting conditions. We propose here a simple method for constructing regression/classification ensembles of neural networks that leads to overtrained aggregate members with an adequate balance between accuracy and diversity. The algorithm is favorably tested against other methods recently proposed in the literature, producing an improvement in performance on the standard statistical databases used as benchmarks. In addition, and as a concrete application, we apply our method to the sunspot time series and predict the remainder of the current cycle 23 of solar activity.

Data Collection↗