A self-organizing incremental neural network for continual supervised learning

Wiwatcharakoses, Chayut; Berrar, Daniel

doi:10.1016/j.eswa.2021.115662

Cited by 19 publications

(10 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In [33], the authors have introduced a new memory population approach (CBRS) for continual online learning that deals with imbalanced and temporally correlated data. Other pertinent methods for enhancing Online Continual Learning were suggested in [34] [35] [36]. For data deduplication, even if the deduplication model is trained with high-quality pairs, features defining duplications may change over time, especially when data is human input.…”

Section: E Continual Learning (Model Retraining)mentioning

confidence: 99%

An End-to-End Big Data Deduplication Framework based on Online Continuous Learning

Elouataoui¹,

Alaoui²,

Mendili³

et al. 2022

IJACSA

View full text Add to dashboard Cite

While big data benefits are numerous, most of the collected data is of poor quality and, therefore, cannot be effectively used as it is. One pre-processing the leading big data quality challenges is data duplication. Indeed, the gathered big data are usually messy and may contain duplicated records. The process of detecting and eliminating duplicated records is known as Deduplication, or Entity Resolution or also Record Linkage. Data deduplication has been widely discussed in the literature, and multiple deduplication approaches were suggested. However, few efforts have been made to address deduplication issues in Big Data Context. Also, the existing big data deduplication approaches are not handling the case of the decreasing performance of the deduplication model during the serving. In addition, most current methods are limited to duplicate detection, which is part of the deduplication process. Therefore, we aim through this paper to propose an End-to-End Big Data Deduplication Framework based on a semi-supervised learning approach that outperforms the existing big data deduplication approaches with an F-score of 98,21%, a Precision of 98,24% and a Recall of 96,48%. Moreover, the suggested framework encompasses all data deduplication phases, including data preprocessing and preparation, automated data labeling, duplicate detection, data cleaning, and an auditing and monitoring phase. This last phase is based on an online continual learning strategy for big data deduplication that allows addressing the decreasing performance of the deduplication model during the serving. The obtained results have shown that the suggested continual learning strategy has increased the model accuracy by 1,16%. Furthermore, we apply the proposed framework to three different datasets and compare its performance against the existing deduplication models. Finally, the results are discussed, conclusions are made, and future work directions are highlighted.

show abstract

Section: E Continual Learning (Model Retraining)mentioning

confidence: 99%

An End-to-End Big Data Deduplication Framework based on Online Continuous Learning

Elouataoui¹,

Alaoui²,

Mendili³

et al. 2022

IJACSA

View full text Add to dashboard Cite

show abstract

“…Typical classifiers of this type are Episodic-GWR [38] and ASOINN Classifier (ASC) [8], which utilize GWR and ASOINN, respectively. One state-of-the-art algorithm is SOINN+ with ghost nodes (GSOINN+) [10]. GSOINN+ has successfully improved the classification performance by generating some ghost nodes near a decision boundary of each class.…”

Section: B Classification Algorithms Capable Of Continual Learningmentioning

confidence: 99%

“…This section presents quantitative comparisons for classification performance of ASC [8], FTCAC, SOINN+C, GSOINN+ [10], CAEAC, CAEAC-I, and CAEAC-C. The source code of ASC 2 , FTCA 3 , SOINN+ 4 , and GSOINN+ 5 is provided by the authors of the related papers.…”

Section: Simulation Experimentsmentioning

confidence: 99%

“…Self-Organizing Incremental Neural Networks (SOINN) [6] is a well-known growing self-organizing clustering algorithm which is inspired by Growing Neural Gas (GNG) [7]. Several conventional studies have shown that SOINN-based classifiers capable of continual learning have good classification performance [8]- [10]. However, the self-organizing process of SOINN-based algorithms is highly unstable due to the instability of GNGbased algorithms.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Class-wise Classifier Design Capable of Continual Learning using Adaptive Resonance Theory-based Topological Clustering

Masuyama¹,

Tsubota²,

Nojima³

et al. 2022

Preprint

View full text Add to dashboard Cite

This paper proposes a supervised classification algorithm capable of continual learning by utilizing an Adaptive Resonance Theory (ART)-based growing self-organizing clustering algorithm. The ART-based clustering algorithm is theoretically capable of continual learning, and the proposed algorithm independently applies it to each class of training data for generating classifiers. Whenever an additional training data set from a new class is given, a new ART-based clustering will be defined in a different learning space. Thanks to the above-mentioned features, the proposed algorithm realizes continual learning capability. Simulation experiments showed that the proposed algorithm has the superior classification performance compared with state-of-the-art clustering-based classification algorithms capable of continual learning.

show abstract

“…As a GNG-based algorithm, Grow When Required (GWR) [18] and gamma-GWR [19] are successful algorithms which appropriately calculate a similarity threshold to prevent an excessive node creation. As a SOINN-based algorithm, SOINN+ [20] and SOINN+ with ghost nodes (GSOINN+) [21] can detect clusters of arbitrary shapes in noisy data streams while avoiding catastrophic forgetting. Another successful approach is ART-based clustering algorithms such as Fuzzy ART [7], Bayes ART [8], and their variants [9], [10].…”

Section: Literature Review a Clustering Algorithms Capable Of Continu...mentioning

confidence: 99%

Adaptive Resonance Theory-Based Topological Clustering With a Divisive Hierarchical Structure Capable of Continual Learning

Masuyama¹,

Amako

Yuna

et al. 2022

IEEE Access

View full text Add to dashboard Cite

Adaptive Resonance Theory (ART) is considered as an effective approach for realizing continual learning thanks to its ability to handle the plasticity-stability dilemma. In general, however, the clustering performance of ART-based algorithms strongly depends on the specification of a similarity threshold, i.e., a vigilance parameter, which is data-dependent and specified by hand. This paper proposes an ART-based topological clustering algorithm with a mechanism that automatically estimates a similarity threshold from the distribution of data points. In addition, for improving information extraction performance, a divisive hierarchical clustering algorithm capable of continual learning is proposed by introducing a hierarchical structure to the proposed algorithm. Experimental results demonstrate that the proposed algorithm has high clustering performance comparable with recently-proposed state-of-the-art hierarchical clustering algorithms.

show abstract

A self-organizing incremental neural network for continual supervised learning

Cited by 19 publications

References 22 publications

An End-to-End Big Data Deduplication Framework based on Online Continuous Learning

An End-to-End Big Data Deduplication Framework based on Online Continuous Learning

Class-wise Classifier Design Capable of Continual Learning using Adaptive Resonance Theory-based Topological Clustering

Adaptive Resonance Theory-Based Topological Clustering With a Divisive Hierarchical Structure Capable of Continual Learning

Contact Info

Product

Resources

About