An approach to reservoir computing design and training

Ferreira, Aida A.; Ludermir, Teresa B.; Aquino, Ronaldo R. B. de

doi:10.1016/j.eswa.2013.01.029

Cited by 65 publications

(39 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…The CSELM method first searches an optimal number of cluster using an optimizing algorithm, JADE, and subsequently determine an ELM for each cluster found. The performance of the CSELM was assessed using five distinct datasets using three types of errors, and was compared to the forecasts generated by different types of networks [14,20,33,34,36]. CSELM is faster than others methods while attaining similar or even better performance.…”

Section: Discussionmentioning

confidence: 99%

An automatic method for construction of ensembles to time series prediction

Lima

Ludermir

2013

HIS

Self Cite

View full text Add to dashboard Cite

We present here a work that applies an automatic construction of ensembles based on the Clustering and Selection (CS) algorithm for time series forecasting. The automatic method, called CSELM, initially finds an optimum number of clusters for training data set and subsequently designates an Extreme Learning Machine (ELM) for each cluster found. For model evaluation, the testing data set are submitted to clustering technique and the nearest cluster to data input will give a supervised response through its associated ELM. Self-organizing maps were used in the clustering phase. Adaptive differential evolution was used to optimize the parameters and performance of the different techniques used in the clustering and prediction phases. The results obtained with the CSELM method are compared with results obtained by other methods in the literature. Five well-known time series were used to validate CSELM.

show abstract

Section: Discussionmentioning

confidence: 99%

An automatic method for construction of ensembles to time series prediction

Lima

Ludermir

2013

HIS

Self Cite

View full text Add to dashboard Cite

show abstract

“…To scale the parameters is necessary to compute the spectral radius of the reservoir matrix. The computation of the spectra requires an important computational effort [11]. Some attempts to generate a procedure for initializing the RC models were introduced in [11], [25], [29]- [31].…”

Section: A Formalization Of the Echo State Network Modelmentioning

confidence: 99%

“…As a consequence, the setting procedure can be expensive in computational time. For instance, the time complexity of an algorithm that computes the spectral radius of a N × N matrix is equal to O(N 4 ) [11].…”

Section: Introductionmentioning

confidence: 99%

An empirical study of L<inf>2</inf>-Boost with Echo State Networks

Basterrech

2013

2013 13th International Conference on Intellient Systems Design and Applications

View full text Add to dashboard Cite

A particular case of Recurrent Neural Network (RNN) was introduced at the beginning of the 2000s under the name of Echo State Networks (ESNs). The ESN model overcomes the limitations during the training of the RNNs while introducing no significant disadvantages. Although the model presents some well-identified drawbacks when the parameters are not well initialized. The performance of an ESN is highly dependent on its internal parameters and pattern of connectivity of the hiddenhidden weights Often, the tuning of the network parameters can be hard and can impact in the accuracy of the models.In this work, we investigate the performance of a specific boosting technique (called L2-Boost) with ESNs as single predictors. The L2-Boost technique has been shown to be an effective tool to combine "weak" predictors in regression problems. In this study, we use an ensemble of random initialized ESNs (without control their parameters) as "weak" predictors of the boosting procedure. We evaluate our approach on five well-know time-series benchmark problems. Additionally, we compare this technique with a baseline approach that consists of averaging the prediction of an ensemble of ESNs.

show abstract

“…They focused on optimising different features of an ESN architecture, including its weights [32], topology [33], design parameters [34] or combinations of them [35]. Applying EAs to optimize ESN designs have shown to produce better network performances than those with recommended settings [36].…”

Section: Optimising Esn Design Parameters and Weightsmentioning

confidence: 99%

Learning Structures of Conceptual Models from Observed Dynamics Using Evolutionary Echo State Networks

Abdelbari

Shafi

2017

Journal of Artificial Intelligence and Soft Computing Research

View full text Add to dashboard Cite

Conceptual or explanatory models are a key element in the process of complex system modelling. They not only provide an intuitive way for modellers to comprehend and scope the complex phenomena under investigation through an abstract representation but also pave the way for the later development of detailed and higher-resolution simulation models. An evolutionary echo state network-based method for supporting the development of such models, which can help to expedite the generation of alternative models for explaining the underlying phenomena and potentially reduce the manual effort required, is proposed. It relies on a customised echo state neural network for learning sparse conceptual model representations from the observed data. In this paper, three evolutionary algorithms, a genetic algorithm, differential evolution and particle swarm optimisation are applied to optimize the network design in order to improve model learning. The proposed methodology is tested on four examples of problems that represent complex system models in the economic, ecological and physical domains. The empirical analysis shows that the proposed technique can learn models which are both sparse and effective for generating the output that matches the observed behaviour.

show abstract

An approach to reservoir computing design and training

Cited by 65 publications

References 18 publications

An automatic method for construction of ensembles to time series prediction

An automatic method for construction of ensembles to time series prediction

An empirical study of L<inf>2</inf>-Boost with Echo State Networks

Learning Structures of Conceptual Models from Observed Dynamics Using Evolutionary Echo State Networks

Contact Info

Product

Resources

About