Deep learning can extract deep sematic features of histopathological images, which plays an important role in machinelearning based disease diagnosis. However, with the improvement of the network level, the extracted...
详细信息
Deep learning can extract deep sematic features of histopathological images, which plays an important role in machinelearning based disease diagnosis. However, with the improvement of the network level, the extracted image features increase dramatically. To avoid the curse of dimensionality, this work proposes a deep feature dimension reduction method, i.e. discriminant-oriented extreme learning machine autoencoder (DELM-AE), for the histopathological image recognition task. Considering the high computational efficiency of the ELM-AE algorithm, but with inherent limitations in classification tasks, the penalty terms to enhance sample discriminability are added to the objective function of DELM-AE, and then a fast algorithm for determining the projection coefficient matrix is presented. Through comparison experiments on dimensionality reduction representation and recognition of histopathology image deep features extracted based on ResNet50, it was verified that the DELM-AE based method can improve the effectiveness and generalization ability of ML classifiers while reducing feature dimensions on two datasets, demonstrating the potential of its feature representation for classification tasks.
As a single learner, extreme learning machine autoencoder (ELM-AE) and generalized extreme learning machine autoencoder (GELM-AE) have limited ability to learn high-dimensional complex data features because high-dimen...
详细信息
As a single learner, extreme learning machine autoencoder (ELM-AE) and generalized extreme learning machine autoencoder (GELM-AE) have limited ability to learn high-dimensional complex data features because high-dimensional data contains more complex and rich discriminative information. GELM-AE only pays attention to the internal relationship of each data subset for dimensionality reduction, ignoring the relationship between different subsets. This paper proposes the homogeneous ensemble extreme learning machine autoencoder (HeELM-AE) to extract high-dimensional complex data diversity features. This method combines the ideas of ensemble feature learning and mutual representation matrix learning. Multiple data subsets are constructed from the original high-dimensional complex dataset with feature learning methods. Generalized extreme learning machine autoencoder(GELM-AE) is used as a base dimension reducer to learn rich discriminative information from highly redundant features. Mutual representation learning methods can characterize correlations between different data subsets and the local manifold structure inherent in different data subsets is maintained through manifold regularization at the same time. Extensive comparative experiments on medical datasets show that compared with other ensemble feature learning models, HeELM-AE is an efficient and accurate model. Finally, visual analysis is used to explain the working mechanism of each stage of HeELM-AE and explore feature learning model interpretability.
Device-free localization (DFL) is becoming one of the new techniques in wireless localization field, due to its advantage that the target to be localized does not need to attach any electronic device. One of the key i...
详细信息
Device-free localization (DFL) is becoming one of the new techniques in wireless localization field, due to its advantage that the target to be localized does not need to attach any electronic device. One of the key issues of DFL is how to characterize the influence of the target on the wireless links, such that the target's location can be accurately estimated by analyzing the changes of the signals of the links. Most of the existing related research works usually extract the useful information from the links through manual approaches, which are labor-intensive and time-consuming. Deep learning approaches have attempted to automatically extract the useful information from the links, but the training of the conventional deep learning approaches are time-consuming, because a large number of parameters need to be fine-tuned multiple times. Motivated by the fast learning speed and excellent generalization performance of extremelearningmachine (ELM), which is an emerging training approach for generalized single hidden layer feed-forward neural networks (SLFNs), this paper proposes a novel hierarchical ELM based on deep learning theory, named multilayer probability ELM (MP-ELM), for automatically extracting the useful information from the links, and implementing fast and accurate DFL. The proposed MP-ELM is stacked by ELM autoencoders, so it also keeps the very fast learning speed of ELM. In addition, considering the uncertainty and redundant links existing in DFL, MP-ELM outputs the probabilistic estimation of the target's location instead of the deterministic output. The validity of the proposed MP-ELM-based DFL is evaluated both in the indoor and the outdoor environments, respectively. Experimental results demonstrate that the proposed MP-ELM can obtain better performance compared with classic ELM, multilayer ELM (ML-ELM), hierarchical ELM (H-ELM), deep belief network (DBN), and deep Boltzmann machine (DBM). (C) 2019 Elsevier B.V. All rights reserved.
In many practical applications, the data are class imbalanced. Accordingly, it is very meaningful and valuable to investigate the classification of imbalanced data. In the framework of binary imbalanced data classific...
详细信息
In many practical applications, the data are class imbalanced. Accordingly, it is very meaningful and valuable to investigate the classification of imbalanced data. In the framework of binary imbalanced data classification, the synthetic minority oversampling technique (SMOTE) is the best-known oversampling method. However, for each positive sample, SMOTE generates only k synthetic samples on the lines between the positive sam-ple and its k-nearest neighbors, resulting in three drawbacks: (1) SMOTE cannot effectively extend the training field of positive samples;(2) the generated positive samples lack diver-sity;(3) SMOTE does not accurately approximate the probability distribution of the posi-tive samples. Therefore, two binary imbalanced data classification methods named BIDC1 and BIDC2 based on diversity oversampling by generative models are proposed. The BIDC1 and BIDC2 conduct diversity oversampling using extreme learning machine autoencoder and generative adversarial network, respectively. Extensive experiments on 26 data sets are conducted to compare the two methods with 14 state-of-the-art methods using five metrics: F-measure, G-means, AUC-area, MMD-score, and Silhouette-score. The experimental results demonstrate that the two methods outperform the other 14 methods. (c) 2021 Elsevier Inc. All rights reserved.
Intelligent fault diagnosis based on deep learning (DL) has been widely used in various engineering practices. However, when confronting massive unlabeled industrial data, traditional data-driven intelligent fault dia...
详细信息
Intelligent fault diagnosis based on deep learning (DL) has been widely used in various engineering practices. However, when confronting massive unlabeled industrial data, traditional data-driven intelligent fault diagnosis approaches cannot fully mine the correlation geometric structure information between data, and therefore cannot obtain good fault diagnosis results. To overcome this difficulty, an efficient fault diagnosis method based on deep hypergraph autoencoder embedding (DHAEE) is presented in this study. First, unlabeled vibration signals are converted into hypergraphs by applying the designed hypergraph construction method. Second, a hypergraph convolutional extreme learning machine autoencoder (HCELM-AE) is designed, which can mine the higher-order structural information and subspace structural information of the original unlabeled data by designing hypergraph convolu-tional and self-representation layers. Furthermore, by stacking multiple HCELM-AE modules in a DL framework, the DHAEE and its corresponding fault diagnosis method is constructed, which not only has the advantage of high computational efficiency of ELM-AE, but also has strong representational learning ability of DL methods. Finally, the effectiveness of the DHAEE based fault diagnosis method is verified by rolling bearing fault data and rotor fault data. Experimental results show that the presented fault diagnosis method has higher accuracy and lower computational complexity than other comparison methods, thus proving that DHAEE is an efficient intelligent massive unlabeled data processing approach for rotating machinery fault diagnosis.(c) 2022 Elsevier B.V. All rights reserved.
Hyperspectral image (HSI) consists of hundreds of narrow spectral band components with rich spectral and spatial information. extremelearningmachine (ELM) has been widely used for HSI analysis. However, the classica...
详细信息
Hyperspectral image (HSI) consists of hundreds of narrow spectral band components with rich spectral and spatial information. extremelearningmachine (ELM) has been widely used for HSI analysis. However, the classical ELM is difficult to use for sparse feature leaning due to its randomly generated hidden layer. In this paper, we propose a novel unsupervised sparse feature learning approach, called Evolutionary Multiobjective-based ELM (EMO-ELM), and apply it to HSI feature extraction. Specifically, we represent the task of constructing the ELM autoencoder (ELM-AE) as a multiobjective optimization problem that takes the sparsity of hidden layer outputs and the reconstruction error as two conflicting objectives. Then, we adopt an Evolutionary Multiobjective Optimization (EMO) method to solve the two objectives, simultaneously. To find the best solution from the Pareto solution set and construct the best trade-off feature extractor, a curvature-based method is proposed to focus on the knee area of the Pareto solutions. Benefited from the EMO, the proposed EMO-ELM is less prone to fall into a local minimum and has fewer trainable parameters than gradient-based AEs. Experiments on two real HSIs demonstrate that the features learned by EMO-ELM not only preserve better sparsity but also achieve superior separability than many existing feature learning methods.
暂无评论