A Novel Broad Learning Model-Based Semi-Supervised Image Classification Method

Abstract

Broad learning system (BLS) is an effective and efficient incremental learning system without the deep architecture. It has strong feature extraction ability and high computational efficiency. However, it is greatly limited in the applicability of supervised learning. For the collected actual data, more data are unlabeled data and less data are labeled data. To overcome these problems, Fick's law assisted propagation (FLAP) is introduced into the BLS to propose a new semi-supervised classification algorithm, namely FLAP-BLS in this paper. In the FLAP-BLS, the FLAP has the labeled ability from the labeled examples to unlabeled examples, it is used to mark plenty of unlabeled samples by few labeled samples in order to obtain a large number of labeled samples and build the sample data matrix. Then an efficient incremental BLS without deep structure can effectively extract features from large-scale data, it is used to effectively classify the sample matrix. Finally, USPS, MNIST and NORB datasets are selected to validate the effectiveness of the FLAP-BLS. The experiment results show that the FLAP-BLS can effectively classify the few labeled samples and a large of unlabeled samples and obtain classification results with high accuracy, and it has faster classification speed, stronger generalization ability and better stability. The proposed method provides a new method for image classification.

An Efficient Algorithm for the Incremental Broad Learning System by Inverse Cholesky Factorization of a Partitioned Matrix

“…To avoid updatingĀ + k , in the next section we will compute the output weightsW k from the inverse Cholesky factorF k , and extend (28) to updateF k iteratively. Accordingly, we can reduce the computational complexity, since the k × k triangular matrixF k is smaller than the k ×l matrixĀ + k , as can be seen from (25).…”

“…To distinguish abnormities under large-scale pipeline network environments, the detection algorithm based on BLS was applied in [32]. BLS was applied in [29] to propose a broad graph-based robust continuous clustering algorithm to upgrade the robust continuous clustering (RCC), and was also utilized in [28] to develop a new semi-supervised classification algorithm to improve the clustering performance of RCC. Moreover, BLS was applied in [33] to develop an efficient and responsive motor fault diagnostic method, was utilized in [34] for the diagnosis of TPIM (three-phase induction motor) faults, and was also used in [35] to propose a new fault diagnosis (PAB-SFD) method for rotor system, which utilizes the principal component analysis (PCA).…”

Calculating Screen Time of Characters in a Video Using Convolutional Neural Networks

“…Broad learning systems are special type of learning systems without deep architecture. Semi supervised learning with effective feature extraction [3] can be implemented in image classification. CNN model which is used in this paper can be applied for image classification including data augmentation [4] process for training the model.…”

An improved fuzzy ARTMAP and Q-learning agent model for pattern classification

“…4) SSL-ART vs. other single SSL models: This experiment compares the performance between SSL-ART with OtM mapping and those of broad learning system (BLS)-based SSL models reported in [52] using three large-scale data sets, i.e., USPS, NORB, and MNIST data sets. Following the same experimental procedure in [52], a total of 9298 samples are randomly selected from each data set, in which 9000 and 298 are training (8500 unlabeled and 500 labeled) and test samples, respectively.…”

“…4) SSL-ART vs. other single SSL models: This experiment compares the performance between SSL-ART with OtM mapping and those of broad learning system (BLS)-based SSL models reported in [52] using three large-scale data sets, i.e., USPS, NORB, and MNIST data sets. Following the same experimental procedure in [52], a total of 9298 samples are randomly selected from each data set, in which 9000 and 298 are training (8500 unlabeled and 500 labeled) and test samples, respectively. SSL-ART produces better results for the NORB and MNIST data sets, while the FLAP-BLS model TABLE III: Accuracy rates ±standard deviations of WESSL-ART, VESSL-ART, and SEMIB and REGB with three base classifiers for binary classification problems, along with numbers of win (w), tie (t) and loss (l).…”