4 results on '"Pérez Sánchez, Beatriz"'
Search Results
2. A review of adaptive online learning for artificial neural networks.
- Author
-
Pérez-Sánchez, Beatriz, Fontenla-Romero, Oscar, and Guijarro-Berdiñas, Bertha
- Subjects
DISTANCE education ,ARTIFICIAL neural networks ,SEQUENTIAL learning ,PROBLEM solving ,DATA analysis - Abstract
In real applications learning algorithms have to address several issues such as, huge amount of data, samples which arrive continuously and underlying data generation processes that evolve over time. Classical learning is not always appropriate to work in these environments since independent and indentically distributed data are assumed. Taking into account the requirements of the learning process, systems should be able to modify both their structures and their parameters. In this survey, our aim is to review the developed methodologies for adaptive learning with artificial neural networks, analyzing the strategies that have been traditionally applied over the years. We focus on sequential learning, the handling of the concept drift problem and the determination of the network structure. Despite the research in this field, there are currently no standard methods to deal with these environments and diverse issues remain an open problem. [ABSTRACT FROM AUTHOR]
- Published
- 2018
- Full Text
- View/download PDF
3. Algoritmos eficientes, incrementales y escalables para el aprendizaje en redes de neuronas artificiales
- Author
-
Pérez-Sánchez, Beatriz, Fontenla-Romero, Óscar, Guijarro-Berdiñas, Bertha, Universidade da Coruña. Departamento de Computación, Fontenla Romero, Óscar, and Guijarro Berdiñas, Bertha
- Subjects
Artificial neural networks ,Redes de neuronas artificiais ,Aprendizaje online ,Concept drift ,Aprendizaje incremental ,Topología adaptativa ,Redes de neuronas arti ciais ,Online learning ,Aprendizaxe online ,Regularization ,Topoloxía adaptativa ,Redes de neuronas artificiales ,Aprendizaje supervisado ,Supervised learning ,Adaptive topology ,Aprendizaxe supervisada ,Regularización ,Cambio no concepto a aprender ,Incremental learning ,Aprendizaxe incremental ,Cambio en el concepto a aprender - Abstract
[Abstract] The core of this work is the development of new supervised learning methods for feedforward neural networks. In the first place, improvements on already developed learning methods are presented in order to generalize their behaviour to new situations while keeping their original characteristics. In particular, the employment of the regularization technique is proposed to handle situations where the overfitting to data is possible. Secondly, we develop online learning algorithms for one and two layer neural networks so as to allow their application in stationary and non stationary contexts in which the process to be modelled does not remain unalterable. In addition, for the case of two layers neural networks, this algorithm will also automatically adapt the topology of the network according to the needs of learning, by adding new hidden units only when necessary. In all cases, an optimum employment of the computational resources is pursued. For all the proposed algorithms, theoretical descriptions of their capacities are included, and their behaviours are illustrated by means of their application to significant cases. Finally, we analyze the results obtained, extracting the main conclusions about each method, their capacities and limitations for their future application. [Resumo] Este traballo céntrase no desenvolvemento de novos modelos de aprendizaxe supervisada para redes de neuronas artificiais alimentadas cara a adiante. En primeiro lugar, expóñense melloras sobre métodos de aprendizaxe xa desenvolvidos, co obxecto de xeneralizar o seu comportamento a novas situacións mantendo as súas características orixinais. En concreto, móstrase o emprego da regularización para manexar situacións onde é posible o fenómeno de sobreaxuste aos datos. En segundo lugar, desenvólvense algoritmos de aprendizaxe online tanto para redes dunha capa como de dúas, que permiten a súa aplicación en contornas non estacionarias en que o proceso que se vai modelar non permanece inalterable. Alén diso, para o caso das redes de dúas capas, este algoritmo online permitirá que tamén a topoloxía da rede se adapte de xeito automático segundo as necesidades da aprendizaxe, engadindo unidades ocultas unicamente en caso de que sexa necesario. En todo momento perséguese un aproveitamento dos recursos dispoñibles. Para os algoritmos propostos, inclúese unha descrición teórica das súas capacidades e o seu comportamento ilústrase mediante a súa aplicación a casos concretos e significativos. Finalmente, analízanse os resultados obtidos, extraendo as principais conclusións do comportamento de cada método e as súas capacidades e limitacións para a súa aplicación futura. [Resumen] Este trabajo se centra en el desarrollo de nuevos modelos de aprendizaje supervisado para redes de neuronas artificiales alimentadas hacia delante. En primer lugar, se plantean mejoras sobre métodos de aprendizaje ya desarrollados, con el objeto de generalizar su comportamiento a nuevas situaciones manteniendo sus características originales. En concreto, se plantea el empleo de la regularización para manejar situaciones donde es posible el fenómeno de sobreajuste a los datos. En segundo lugar, se desarrollan algoritmos de aprendizaje online tanto para redes de una capa como de dos, que permiten su aplicación en entornos no estacionarios en los que el proceso a modelar no permanece inalterable. Además, para el caso de las redes de dos capas, este algoritmo online permitirá que también la topología de la red se adapte de manera autom ática según las necesidades del aprendizaje, añadiendo unidades ocultas únicamente en caso necesario. En todo momento se persigue un aprovechamiento de los recursos disponibles. Para los algoritmos propuestos se incluye una descripción teórica de sus capacidades, y su comportamiento se ilustra mediante su aplicación a casos concretos y significativos. Finalmente se analizan los resultados obtenidos, extrayendo las principales conclusiones del comportamiento de cada método, capacidades y limitaciones para su aplicación futura.
- Published
- 2010
4. A robust incremental learning method for non-stationary environments
- Author
-
Martínez-Rego, David, Pérez-Sánchez, Beatriz, Fontenla-Romero, Oscar, and Alonso-Betanzos, Amparo
- Subjects
- *
MACHINE learning , *ARTIFICIAL neural networks , *ROBUST control , *ALGORITHMS , *REGRESSION analysis , *SET theory - Abstract
Abstract: Recent machine learning challenges require the capability of learning in non-stationary environments. These challenges imply the development of new algorithms that are able to deal with changes in the underlying problem to be learnt. These changes can be gradual or trend changes, abrupt changes and recurring contexts. As the dynamics of the changes can be very different, existing machine learning algorithms exhibit difficulties to cope with them. Several methods using, for instance, ensembles or variable length windowing have been proposed to approach this task. In this work we propose a new method, for single-layer neural networks, that is based on the introduction of a forgetting function in an incremental online learning algorithm. This forgetting function gives a monotonically increasing importance to new data. Due to the combination of incremental learning and increasing importance assignment the network forgets rapidly in the presence of changes while maintaining a stable behavior when the context is stationary. The performance of the method has been tested over several regression and classification problems and its results compared with those of previous works. The proposed algorithm has demonstrated high adaptation to changes while maintaining a low consumption of computational resources. [Copyright &y& Elsevier]
- Published
- 2011
- Full Text
- View/download PDF
Catalog
Discovery Service for Jio Institute Digital Library
For full access to our library's resources, please sign in.