Modular neural network architectures for classification

Auda, G.; Kamel, Mohamed S.; Raafat, Hazem M.

doi:10.1109/icnn.1996.549082

Cited by 21 publications

(18 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Figure 4 shows the pattern distribution of the training set and the real boundary between two classes. In back-propagation NNs, the training objective is to minimize the error E (Of course, early stopping criteria [19][20][21] is used to prevent from overtraining), instead of finding (or approaching) the real border between two classes. Here…”

Section: Explaining Why Rpt-hicl Is Better Than Original Hiclmentioning

confidence: 99%

“…The NN classifiers have some merits over traditional pattern recognition systems with respect to the capability of adaptive learning, generalization ability with noisy or sparse learning data, and feasibility for hardware implementation [1]. Current NN classifiers suffer from major drawback of high internal interference because of the strong coupling among their hidden-layer weights [2].…”

Section: Introductionmentioning

confidence: 99%

“…It generally decomposes large-sized tasks into several sub-tasks; each one is handled by a simple, fast, and efficient module. Then, sub-solutions are integrated via a multi-module decision making strategy [1]. Many task decomposition methods have been proposed in the literature [4,[6][7][8][9][10][11][12][13][14][15][16][17][18].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Hierarchical Incremental Class Learning with Reduced Pattern Training

Guan

Bao

Sun

2006

Neural Process Lett

View full text Add to dashboard Cite

Hierarchical Incremental Class Learning (HICL) is a new task decomposition method that addresses the pattern classification problem. The HICL is proven to be a good classifier but closer examination reveals areas for potential improvement. This paper proposes a theoretical model to evaluate the performance of HICL and presents an approach to improve the classification accuracy of HICL by applying the concept of Reduced Pattern Training (RPT). The theoretical analysis shows that HICL can achieve better classification accuracy than Output Parallelism [Guan and Li: IEEE Transaction on Neural Networks, 13 (2002), 542-550]. The procedure for RPT is described and compared with the original training procedure. The RPT reduces systematically the size of the training data set based on the order of sub-networks built. The results from four benchmark classification problems show much promise for the improved model.

show abstract

Section: Explaining Why Rpt-hicl Is Better Than Original Hiclmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Hierarchical Incremental Class Learning with Reduced Pattern Training

Guan

Bao

Sun

2006

Neural Process Lett

View full text Add to dashboard Cite

show abstract

“…In order to avoid large computational cost and overfitting, a method called early stopping based on validation set is used as the stopping criteria. Please refer to [22] for details.…”

Section: A Electronic Image Files (Optional)mentioning

confidence: 99%

Reduced Pattern Training Based on Task Decomposition Using Pattern Distributor

Guan

Bao

Neo

2007

IEEE Trans. Neural Netw.

View full text Add to dashboard Cite

“…In [28,29] Although constructive learning algorithms can automatically find an optimal combination, they are still suffering from drawbacks such as inefficiency in utilizing network resources as the task (and the network) gets larger, and inability of the current learning schemes to cope with high-complexity tasks [14]. Large networks tend to introduce high internal interference because of the strong coupling among their input-tohidden layer weights [15].…”

Section: Introductionmentioning

confidence: 99%

Output partitioning of neural networks

Guan

Tan

et al. 2005

Neurocomputing

View full text Add to dashboard Cite

Many constructive learning algorithms have been proposed to find an appropriate network structure for a classification problem automatically. Constructive learning algorithms have drawbacks especially when used for complex tasks and modular approaches have been devised to solve these drawbacks. At the same time, parallel training for neural networks with fixed configurations has also been proposed to accelerate the training process. A new approach that combines advantages of constructive learning and parallelism, output partitioning, is presented in this paper. Classification error is used to guide the proposed incremental-partitioning algorithm, which divides the original dataset into several smaller sub-datasets with distinct classes. Each sub-dataset is then handled in parallel, by a smaller constructively trained sub-network which uses the whole input vector and produces a portion of the final output vector where each class is represented by one unit. Three classification datasets are used to test the validity of this method, and results show that this method reduces the classification test error.

show abstract

Modular neural network architectures for classification

Cited by 21 publications

References 8 publications

Hierarchical Incremental Class Learning with Reduced Pattern Training

Hierarchical Incremental Class Learning with Reduced Pattern Training

Reduced Pattern Training Based on Task Decomposition Using Pattern Distributor

Output partitioning of neural networks

Contact Info

Product

Resources

About