Multitask learning applied to evolving fuzzy-rule-based predictors

Ayres, Amanda O. C.; Zuben, Fernando J. Von

doi:10.1007/s12530-019-09300-w

Cited by 6 publications

(21 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In 2004, Plamen and Filev published one of the pioneer work in the framework of Evolving Fuzzy Systems [35], in which they discussed an incremental approach for learning Takagi-Sugeno fuzzy rule-based systems. Along the years, several extensions and applications to different fields have interested evolving fuzzy rule-based systems [36,37].…”

Section: Related Workmentioning

confidence: 99%

“…Specifically, we show the results for the chunks which are characterized by the most evident concept drift. Figures [34][35][36][37][38][39] show the trends of the F-measure of the relevant class, for both FHDT and HDT, calculated on Test 1 and Test 2 datasets, respectively. We recall that the trends of the complexity of the models are the same as the ones shown in Figures 30-32.…”

Section: Then Room Is Not Occupiedmentioning

confidence: 99%

See 1 more Smart Citation

Fuzzy Hoeffding Decision Tree for Data Stream Classification

Ducange¹,

Marcelloni²,

Pecori³

2021

IJCIS

View full text Add to dashboard Cite

Data stream mining has recently grown in popularity, thanks to an increasing number of applications which need continuous and fast analysis of streaming data. Such data are generally produced in application domains that require immediate reactions with strict temporal constraints. These particular characteristics make problematic the use of classical machine learning algorithms for mining knowledge from these fast data streams and call for appropriate techniques. In this paper, based on the well-known Hoeffding Decision Tree (HDT) for streaming data classification, we introduce FHDT, a fuzzy HDT that extends HDT with fuzziness, thus making HDT more robust to noisy and vague data. We tested FHDT on three synthetic datasets, usually adopted for analyzing concept drifts in data stream classification, and two real-world datasets, already exploited in some recent researches on fuzzy systems for streaming data. We show that FHDT outperforms HDT, especially in presence of concept drift. Furthermore, FHDT is characterized by a high level of interpretability, thanks to the linguistic rules that can be extracted from it.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Then Room Is Not Occupiedmentioning

confidence: 99%

Fuzzy Hoeffding Decision Tree for Data Stream Classification

Ducange¹,

Marcelloni²,

Pecori³

2021

IJCIS

View full text Add to dashboard Cite

show abstract

“…Evolving systems can change the general structure of the model designed to describe the data stream by updating the data at every time. This change was done by implemented several mechanisms [6].…”

Section: Introductionmentioning

confidence: 99%

Data Stream Clustering Using Fuzzy-based Evolving Cauchy Algorithm

Al-Khamees¹

2021

IJIES

View full text Add to dashboard Cite

Many different applications in the real world can generate huge amount of data, that has unconventional features including massive size, fast access, besides the evolving in its nature; this is data stream. Data stream clustering algorithms began to grow at breakneck speed. evolving Cauchy (eCauchy) is a significant algorithm of density-based data stream clustering. The major limitation of eCauchy is the high number of clusters generated in dynamic environments. This paper presents an evolving model for data stream by optimizing e-Cauchy algorithm to decrease the number of clusters and reach to an ideal number by implementing evolving mechanisms (adding, merging, splitting clusters) based on a specific membership function. Model is tested by two real datasets NSL-KDD99 and keystroke. Proposed model outperforms two other algorithms, e-Cauchy and FEAC-Stream. Model constructs five and four clusters with less time to implement 1.30 and 2.30 minutes respectively for each dataset.

show abstract

“…Furthermore, the distribution of the input data may change dynamically over time, requiring not only the parameters but also the structure of the model to be adaptable along time. Besides these challenges, the online learning task should generally be implemented based on a limited amount of computational resources (Ayres and Von Zuben, 2021b).…”

Section: Introductionmentioning

confidence: 99%

“…Particularly, the evolving extension of the original Fuzzy-Rule-Based (FRB) system (Pedrycz and Gomide, 2007;Kasabov, 2007), denoted evolving Fuzzy-Rule-Based (eFRB) system, divides the input space into fuzzy regions, called information granules, where IF-THEN rules are conceived to provide local predictions to the input sample. The final output predicted by the model is calculated by the weighted average of the local affine models operating as the Takagi-Sugeno (TS) consequent part of the rules (Ayres and Von Zuben, 2021b). The learning mechanism for updating the information granules, each one acting as the antecedent of a fuzzy rule, is executed online in response to the input data stream.…”

Section: Introductionmentioning

confidence: 99%

Evolving multitask models for machine learning

Ortega de Castro Ayres

View full text Add to dashboard Cite

Esta tese de doutorado oferece contribuições para os sistemas evolutivos, uma área de Aprendizado de Máquina em que a estrutura do modelo é construída do zero e adaptada em tempo real de acordo com os dados de entrada e saída que chegam continuamente ao longo do tempo. Uma das mais proeminentes classes de sistemas evolutivos, os sistemas evolutivos baseados em regras fuzzy, automática e recursivamente dividem o espaço de entrada em regiões fuzzy, chamadas de grânulos de informação, onde regras do tipo SE -ENTÃO são definidas para prover predições locais para a amostra de entrada. Com relação aos antecedentes das regras, introduz-se um novo algoritmo chamado Preditor evolutivo de Valores Extremos (Extreme Value evolving Predictor -EVeP na sigla em Inglês), que oferece uma abordagem bem fundamentada para definir os grânulos fuzzy evolutivos que formam os antecedentes e, também, os consequentes das regras. Esses grânulos correspondem à função de inclusão radial Weibull, interpretada pela Teoria dos Valores Extremos como a distribuição limite da proximidade relativa entre as regras do modelo de aprendizado. Com respeito aos consequentes das regras, propõe-se uma nova abordagem que aplica Aprendizado Multitarefa para determinar os parâmetros Takagi-Sugeno usando a intersecção dos grânulos de informação para formar um grafo esparso dos relacionamentos estruturais entre as regras. Assim, diferentemente do mecanismo tradicional empregado na literatura, onde cada regra é responsável pelo seu próprio aprendizado, os termos dos consequentes das regras correspondem a modelos de regressão linear que são calculados pela solução de um problema de aprendizado multitarefa. Por último, uma extensão dessa abordagem para lidar com séries temporais multivariadas é apresentada. Levando-se em conta que cada regra fuzzy é parte de um preditor para uma série temporal específica, introduz-se um procedimento unificado que atualiza o grafo esparso por meio da composição dos relacionamentos par-a-par estabelecidos pelas regras fuzzy com o grau de similaridade de suas correspondentes séries temporais. Com esta abordagem, explora-se não apenas a interação atual das várias regras que compõem cada preditor evolutivo, mas também a interação das múltiplas séries temporais que estão sendo simultaneamente preditas. Uma série de experimentos computacionais foi conduzida, evidenciando um ganho de desempenho das múltiplas abordagens introduzidas nesta tese quando comparadas a vários algoritmos estado-da-arte na literatura.

show abstract

Multitask learning applied to evolving fuzzy-rule-based predictors

Cited by 6 publications

References 27 publications

Fuzzy Hoeffding Decision Tree for Data Stream Classification

Fuzzy Hoeffding Decision Tree for Data Stream Classification

Data Stream Clustering Using Fuzzy-based Evolving Cauchy Algorithm

Evolving multitask models for machine learning

Contact Info

Product

Resources

About