The autoencoder (AE) is a fundamental deep learning approach to anomaly detection. AEs are trained on the assumption that abnormal inputs will produce higher reconstruction errors than normal ones. In practice, howeve...
详细信息
The autoencoder (AE) is a fundamental deep learning approach to anomaly detection. AEs are trained on the assumption that abnormal inputs will produce higher reconstruction errors than normal ones. In practice, however, this assumption is unreliable in the unsupervised case, where the training data may contain anomalous examples. Given sufficient capacity and training time, an AE can generalize to such an extent that it reliably reconstructs anomalies. Consequently, the ability to distinguish anomalies via reconstruction errors is diminished. We respond to this limitation by introducing three new methods to more reliably train AEs for unsupervised anomaly detection: cumulative error scoring (CES), percentile loss (PL), and early stopping via knee detection. We demonstrate significant improvements over conventional AE training on image, remote-sensing, and cybersecurity datasets.
Objective: In long-term video-monitoring, automatic seizure detection holds great promise as a means to reduce the workload of the epileptologist. A convolutional neural network (CNN) designed to process images of EEG...
详细信息
Objective: In long-term video-monitoring, automatic seizure detection holds great promise as a means to reduce the workload of the epileptologist. A convolutional neural network (CNN) designed to process images of EEG plots demonstrated high performance for seizure detection, but still has room for reducing the false-positive alarm rate. Methods: We combined a CNN that processed images of EEC plots with patient-specific autocncoders (AE) of EEC signals to reduce the false alarms during seizure detection. The AE automatically logged abnormalities, i.e., both seizures and artifacts. Based on seizure logs compiled by expert epileptologists and errors made by AE, we constructed a CNN with 3 output classes: seizure, non-seizure-but-abnormal, and non-seizure. The accumulative measure of number of consecutive seizure labels was used to issue a seizure alarm. Results: The second-by-second classification performance of AE-CNN was comparable to that of the original CNN. False-positive seizure labels in AF-CNN were more likely interleaved with "non-seizure-but-abnormal" labels than with true-positive seizure labels. Consequently, "non-seizure-but-abnormal" labels interrupted runs of false-positive seizure labels before triggering an alarm. The median false alarm rate with the AE-CNN was reduced to 0.034 h(-1), which was one-fifth of that of the original CNN (0.17 h(-1)). Conclusions: A label of "non-seizure-but-abnormal" offers practical benefits for seizure detection. The modification of a CNN with an AE is worth considering because Alis can automatically assign "non-seizure-but-abnormal" labels in an unsupervised manner with no additional demands on the time of the epileptologist.
Safety, efficiency, and reliability are essential requirements for aero-engines. Timely and accurate diagnosis of engine faults enables effective planning of maintenance operations and reduces downtime. Although tradi...
详细信息
Safety, efficiency, and reliability are essential requirements for aero-engines. Timely and accurate diagnosis of engine faults enables effective planning of maintenance operations and reduces downtime. Although traditional physics-based methods perform well under controlled test bench scenarios, their effectiveness in handling very noisy data and missing values is limited, constraining their utility in real-world settings. To address these gaps, we propose a fusion autoencoder that combines physics-informed and pattern-informed techniques, augmented with a Beta-Variational autoencoder learning backbone to enhance the robustness of the model. Additionally, a novel health index called the piecewise anomaly index is proposed that can detect and classify faults simultaneously. To evaluate the efficacy of the novel framework, we modified the New Commercial Modular Aero-Propulsion System Simulation (N-CMAPSS) dataset to simulate real-world scenarios and conducted experiments. The results show that the proposed method can detect faults earlier than common techniques, while also achieving accurate fault classification and degree determination with the new index.
Skeleton-based human action recognition (HAR) is being utilized in various fields like action classification and abnormal behavior detection. The accurate coordinates of the human joints are a crucial factor for the h...
详细信息
Skeleton-based human action recognition (HAR) is being utilized in various fields like action classification and abnormal behavior detection. The accurate coordinates of the human joints are a crucial factor for the high performance in skeleton-based HAR. However, the missing joints caused by occlusion and invisibility result in performance degradation. Hence, in this paper, a missing joint reconstruction model is proposed to improve the performance of skeleton-based HAR. The proposed model, based on a denoising graph autoencoder (DGAE), regards missing joints as noise corrupted information and aims to reconstruct them to be close to their original coordinates. When the encoder of the proposed model compresses the noised input into a latent vector, a masking Laplacian matrix is introduced to reduce the effect of the missing joints' features. The masking Laplacian matrix adjusts the effect of features between a missing joint and its adjacent joints by altering the weights of an adjacent matrix. In the decoder, a Laplacian matrix, which represents the connections among the joints, is utilized to reconstruct an output from the latent vector. The experiment result shows that the proposed model reconstructs the coordinates of missing joints with a marginal error. In addition, the performance of skeleton-based HAR is enhanced by reconstructing the missing joints.
In this work, we introduce the graph regularized autoencoder. We propose three variants. The first one is the unsupervised version. The second one is tailored for clustering, by incorporating subspace clustering terms...
详细信息
In this work, we introduce the graph regularized autoencoder. We propose three variants. The first one is the unsupervised version. The second one is tailored for clustering, by incorporating subspace clustering terms into the autoencoder formulation. The third is a supervised label consistent autoencoder suitable for single label and multi-label classification problems. Each of these has been compared with the state-of-the-art on benchmark datasets. The problems addressed here are image denoising, clustering and classification. Our proposed methods excel of the existing techniques in all of the problems. (c) 2018 Elsevier Ltd. All rights reserved.
For predicting the value of the quality variable in fermentation processes, traditional data-driven methods do not use information in large amounts of unlabelled data. To solve this data-rich but information-poor (DRI...
详细信息
For predicting the value of the quality variable in fermentation processes, traditional data-driven methods do not use information in large amounts of unlabelled data. To solve this data-rich but information-poor (DRIP) problem, a teacher student stacked sparse recurrent autoencoder (TS-SSRAE) model is proposed. Compared with traditional data-driven methods, the proposed method has three main advantages. First, an autoencoder is an unsupervised method which can effectively extract rich information in unlabelled data. The proposed stacked recurrent autoencoder (SRAE) with long short-term memory (LSTM) recurrent neural unit is superior to traditional autoencoders when extracting the dynamic correlation information in the fermentation process. Second, sparse constraints can make it much easier for hidden neurons to obtain useful information in a single moment. Finally, the LSTM recurrent neural unit is complex and the inputs of a SRAE must be a sequence, which increases the complexity of the model to a certain extent. So, the knowledge distillation is employed to simplify the model and reduce the computing time. In order to demonstrate its effectiveness, the proposed method is applied to the penicillin fermentation process for a simulation experiment and Escherichia coli production of interleukin-2. The results show that the proposed method based on TS-SSRAE can have better performance than conventional methods.
Rolling bearing is a critical component of machinery that has been widely applied in manufacturing, transportation, aerospace, and power and energy industries. The timely and accurate bearing fault detection thus is o...
详细信息
Rolling bearing is a critical component of machinery that has been widely applied in manufacturing, transportation, aerospace, and power and energy industries. The timely and accurate bearing fault detection thus is of vital importance. Computational data-driven deep learning has recently become a prevailing approach for bearing fault detection. Despite the progress of the deep learning approach, the deep learning performance is hinged upon the size of labeled data, the acquisition of which is expensive in actual implementation. Unlabeled data, on the other hand, are inexpensive. In this research, we develop a new semi-supervised learning method built upon the autoencoder to fully utilize a large amount of unlabeled data together with limited labeled data to enhance fault detection performance. Compared with the state-of-the-art semi-supervised learning methods, this proposed method can be more conveniently implemented with fewer hyperparameters to be tuned. In this method, a joint loss is established to account for the effects of labeled and unlabeled data, which is subsequently used to direct the backpropagation training. Systematic case studies using the Case Western Reserve University (CWRU) rolling bearing dataset are carried out, in which the effectiveness of this new method is verified by comparing it with other well-established baseline methods. Specifically, nearly all emulation runs using the proposed methodology can lead to around 2%-5% accuracy increase, indicating its robustness in performance enhancement.
Detecting anomalies such as breakage and excessive wear of cutting tools in the machining process is crucial to prevent damage and improve productivity. Data-driven anomaly detection (AD) methods suffer from limited a...
详细信息
Detecting anomalies such as breakage and excessive wear of cutting tools in the machining process is crucial to prevent damage and improve productivity. Data-driven anomaly detection (AD) methods suffer from limited availability of anomaly samples, which is ineluctable in practice owing to strict reliability restrictions. Therefore, we propose a semisupervised AD approach in which only failure-free samples are required to establish an AD model. The key strategy is to learn the characteristics of failure-free samples using an improved autoencoder (AE) and discern observations by deviations from the characteristics. We rebuild the loss function of AE to impel the model to learn the common characteristics in latent space. We propose a factor that reflects the anomaly degree as the decision-making function to implement AD. The proposed approach is verified on an experimental cutting tool breakage dataset and a public cutting tool wear dataset. The experimental results demonstrate the validity of the proposed approach. The comparisons with conventional methods substantiate that the proposed approach outperforms existing AD methods.
Deep learning has been developed to generate promising super resolution hyperspectral imagery by fusing hyperspectral imagery with the panchromatic ***,it is still challenging to maintain edge spectral information in ...
详细信息
Deep learning has been developed to generate promising super resolution hyperspectral imagery by fusing hyperspectral imagery with the panchromatic ***,it is still challenging to maintain edge spectral information in the necessary upsampling processes of these approaches,and diffcult to guarantee effective feature *** study proposes a pansharpening network denoted as HyperRefiner that consists of,(1)a well performing upsampling network SRNet,in which the dual attention block and refined attention block are cascaded to accomplish the extraction and fusion of features;(2)a spectral autoencoder that is embedded to perform dimensionality reduction under constrained feature extraction;and(3)the optimization module which performs self-attention at the pixel and feature levels.A comparisonwithseveral state-of-the-art models reveals that HyperRefiner can improve the quality of the fused ***,compared to the single-head HyperTransformer and with the Chikusei dataset,our network improved the Peak Signal-to-Noise Ratio,Erreur Relative Globale Adimensionnelle de Synthese and Spectral Angle Mapper by 0.86%,3.62%,and 2.09%,and reduce the total memory,floating point operations,model parameters and computation time by 41%,75%,86%and 46%,*** experimental results show that HyperRefiner outperforms several networks and demonstrates its usefulness in hyperspectral image *** code is publicly available athttps://***/zsspo/Fusion_HyperRefiner.
作者:
Zhang, DonglinWu, Xiao-JunJiangnan Univ
Sch Artificial Intelligence & Comp Sci Wuxi 214122 Jiangsu Peoples R China Jiangnan Univ
Jiangsu Prov Engn Lab Pattern Recognit & Computat Wuxi 214122 Jiangsu Peoples R China
Hashing methods have sparked great attention on multimedia tasks due to their effectiveness and efficiency. However, most existing methods generate binary codes by relaxing the binary constraints, which may cause larg...
详细信息
Hashing methods have sparked great attention on multimedia tasks due to their effectiveness and efficiency. However, most existing methods generate binary codes by relaxing the binary constraints, which may cause large quantization error. In addition, most supervised cross-modal approaches preserve the similarity relationship by constructing an n x n large-size similarity matrix, which requires huge computation, making these methods unscalable. To address the above challenges, this article presents a novel algorithm, called scalable discrete matrix factorization and semantic autoencoder method (SDMSA). SDMSA is a two-stage method. In the first stage, the matrix factorization scheme is utilized to learn the latent semantic information, the label matrix is incorporated into the loss function instead of the similarity matrix. Thereafter, the binary codes can be generated by the latent representations. During optimization, we can avoid manipulating a large nxn similarity matrix, and the hash codes can be generated directly. In the second stage, a novel hash function learning scheme based on the autoencoder is proposed. The encoder-decoder paradigm aims to learn projections, the feature vectors are projected to code vectors by encoder, and the code vectors are projected back to the original feature vectors by the decoder. The encoder-decoder scheme ensures the embedding can well preserve both the semantic and feature information. Specifically, two algorithms SDMSA-lin and SDMSA-ker are developed under the SDMSA framework. Owing to the merit of SDMSA, we can get more semantically meaningful binary hash codes. Extensive experiments on several databases show that SDMSA-lin and SDMSA-ker achieve promising performance.
暂无评论