In an object detection system, the main objective during training is to maintain the detection and false positive rates under acceptable levels when the model is run over the test set. However, this typically translat...
详细信息
In an object detection system, the main objective during training is to maintain the detection and false positive rates under acceptable levels when the model is run over the test set. However, this typically translates into an unacceptable rate of false alarms when the system is deployed in a real surveillance scenario. To deal with this situation, which often leads to system shutdown, we propose to add a filter step to discard part of the new false positive detections that are typical of the new scenario. This step consists of a deep autoencoder trained with the false alarm detections generated after running the detector over a period of time in the new scenario. Therefore, this step will be in charge of determining whether the detection is a typical false alarm of that scenario or whether it is something anomalous for the autoencoder and, therefore, a true detection. In order to decide whether a detection must be filtered, three different approaches have been tested. The first one uses the autoencoder reconstruction error measured with the mean squared error to make the decision. The other two use thek-NN (k-nearest neighbors) and one-class SVMs (support vector machines) classifiers trained with the autoencoder vector representation. In addition, a synthetic scenario has been generated with Unreal Engine 4 to test the proposed methods in addition to a dataset with real images. The results obtained show a reduction in the number of false positives between 22.5% and 87.2% and an increase in the system's precision of 1.2%-47% when the autoencoder is applied.
Efficient solvers for traveling salesman problem (TSP) have great significance in the field of consumer electronic systems and devices. Existing studies require independent and repetitive runs for similar TSPs. To uti...
详细信息
Efficient solvers for traveling salesman problem (TSP) have great significance in the field of consumer electronic systems and devices. Existing studies require independent and repetitive runs for similar TSPs. To utilize useful knowledge buried in the twin TSP which is similar to the target TSP, a twin learning framework based on task matching and mapping strategy is proposed. We use an autoencoder to extract the feature vectors of the historical tasks and the target task to find the twin task. If not found, construct a twin task based on a graph-filter. Further, we get the solution of the target task with the optimized twin task by learning a mapping matrix from the twin task to the target task. Finally, we incorporate local search algorithms into the twin learning framework, which called strategy mapping solver (SMS) to further improve the quality of the solution. The efficacy of the SMS is evaluated through comprehensive empirical studies with commonly used TSP benchmarks, against meta-heuristic algorithms and a learning improvement heuristics algorithm, demonstrating its effectiveness for TSP. Moreover, a real-world combinatorial optimization application, laser engraving path planning, is presented to further confirm the efficacy of SMS.
We propose an unsupervised polarimetric synthetic aperture radar (PolSAR) land classification system consisting of a series of two unsupervised neural networks, namely, a quaternion autoencoder and a quaternion self-o...
详细信息
We propose an unsupervised polarimetric synthetic aperture radar (PolSAR) land classification system consisting of a series of two unsupervised neural networks, namely, a quaternion autoencoder and a quaternion self-organizing map (SOM). Most of the existing PolSAR land classification systems use a set of feature information that humans designed beforehand. However, such methods will face limitations in the near future when we expect classification into a large number of land categories recognizable to humans. By using a quaternion autoencoder, our proposed system extracts feature information based on the natural distribution of PolSAR features. In this paper, we confirm that the information necessary for land classification is extracted as the features while noise is filtered. Then, we show that the extracted features are classified by the quaternion SOM in an unsupervised manner. As a result, we can discover even new and more detailed land categories. For example, town areas are divided into residential areas and factory sites, and grass areas are subcategorized into furrowed farmlands and flat grass areas. We also examine the realization of topographic mapping of the features in the SOM space.
In order to speed up the process of optimizing design of metasurface absorbers, an improved design model for metasurface absorbers based on autoencoder (AE) and BiLSTM-Attention-FCN-Net (including bidirectional long-s...
详细信息
In order to speed up the process of optimizing design of metasurface absorbers, an improved design model for metasurface absorbers based on autoencoder (AE) and BiLSTM-Attention-FCN-Net (including bidirectional long-short-term memory network, attention mechanism, and fully-connection layer network) is proposed. The metasurface structural parameters can be input into the forward prediction network to predict the corresponding absorption spectra. Meantime, the metasurface structural parameters can be obtained by inputting the absorption spectra into the inverse prediction network. Specially, in the inverse prediction network, the bidirectional long-short-term memory (BiLSTM) network can effectively capture the context relationship between absorption spectral sequence data, and the attention mechanism can enhance the BiLSTM output sequence features, which highlight the critical feature information. After the training, the mean square error (MSE) value on the validation set of the reverse prediction network converges to 0.0046, R2 reaches 0.975, and our network can accurately predict the metasurface structure parameters within 1.5 s with a maximum error of 0.03 mm. Moreover, this model can achieve the optimal design of multi-band metasurface absorbers, including the single-band, dual-band, and three-band absorptions. The proposed method can also be extended to other types of metasurface optimization design.
Various applications are deployed on mobile smart devices in almost every situations of our life, while in some of these situations sensitive applications are strictly prohibited, such as cameras in cinemas and browse...
详细信息
Various applications are deployed on mobile smart devices in almost every situations of our life, while in some of these situations sensitive applications are strictly prohibited, such as cameras in cinemas and browsers in examination halls. Real-time recognition of applications running on mobile smart devices is of great significance in these cases. However, most of the existing technologies have the limitation that they require system permissions to obtain the running application list which is banned by mainstream mobile operating systems. Noting that the launch of a certain application will emit a unique pattern of magnetic field, we introduce magnetic field side channel analysis to recognize running applications. However, magnetic field side channel analysis is challenging since it is hard to extract features from magnetic field data without domain experts. Besides, real-time applications identification demands accurate detection of applications launching. To overcome these challenges, we extract robust depth features using autoencoder and implement online application recognition by introducing finite-state machine to identify the application launch window from raw data. The proposed method is evaluated by recognizing 1000 different applications in real environment. The experiment results show that the proposed method is feasible and effective in real-time application identification.
This paper proposes a robust autoencoder withWasserstein distance metric to extract the linear separability features from the input data. To minimize the difference between the reconstructed feature space and the orig...
详细信息
This paper proposes a robust autoencoder withWasserstein distance metric to extract the linear separability features from the input data. To minimize the difference between the reconstructed feature space and the original feature space, using Wasserstein distance realizes a homeomorphic transformation of the original feature space, i.e., the so-called the reconstruction of feature space. The autoencoder is used for features extraction of linear separability in the reconstructed feature space. Experiment results on real datasets show that the proposed method reaches up 0.9777 and 0.7112 on the low-dimensional and high-dimensional datasets in extracted accuracies, respectively, and also outperforms competitors. Results also confirm that compared with feature metric-based methods and deep network architectures-based method, the linear separabilities of those features extracted by distance metric-based methods win over them. More importantly, the linear separabilities of those features obtained by evaluating distance similarity of the data are better than those obtained by evaluating feature importance of data. We also demonstrate that the data distribution in the feature space reconstructed by a homeomorphic transformation can be closer to the original data distribution.
autoencoder (AE) has been widely used in multivariate statistical process monitoring (MSPM) and various AE-based methods have been applied in fault detection. Process data is often contaminated by outliers and measure...
详细信息
autoencoder (AE) has been widely used in multivariate statistical process monitoring (MSPM) and various AE-based methods have been applied in fault detection. Process data is often contaminated by outliers and measurement noise which may lead to the overfitting problem for AE-based methods. In this paper, a novel feature extraction method called low-rank reconstruction-based autoencoder (LRAE) is proposed for robust fault detection. LRAE decomposes the input into a combination of a low-rank data matrix and a noise matrix. By penalizing the rank of the data matrix, LRAE separates the low-rank clean data from the contaminated process data. Instead of directly reconstructing the loss between the input data and the output data, we design a low-rank reconstruction strategy, i.e. reconstruct the loss between the low-rank clean data and the output of the AE. The proposed LRAE can be trained end-to-end by jointly optimizing an AE and a low-rank approximation. LRAE is a nonlinear method which can tackle the complicated process data better than the linear methods such as principal component analysis (PCA). Moreover, the optimization of the low-rank approximation provides the robustness of LRAE to reconstruct the clean data in the output layer when the input process data is contaminated. After training, the features of the hidden layer can be computed for further fault detection. Extensive experiments demonstrate that LRAE outperforms traditional fault detection methods, including PCA, robust principal component analysis (RPCA), kernel principal component analysis (KPCA), AE, and denoising autoencoder (DAE). Especially, LRAE provides more robust results when the process data suffer from outliers and measurement noise.
One-class classification has gained interest as a solution to certain kinds of problems typical in a wide variety of real environments like anomaly or novelty detection. autoencoder is the type of neural network that ...
详细信息
One-class classification has gained interest as a solution to certain kinds of problems typical in a wide variety of real environments like anomaly or novelty detection. autoencoder is the type of neural network that has been widely applied in these one-class problems. In the Big Data era, new challenges have arisen, mainly related with the data volume. Another main concern derives from Privacy issues when data is distributed and cannot be shared among locations. These two conditions make many of the classic and brilliant methods not applicable. In this paper, we present distributed singular value decomposition (DSVD-autoencoder), a method for autoencoders that allows learning in distributed scenarios without sharing raw data. Additionally, to guarantee privacy, it is noniterative and hyperparameter-free, two interesting characteristics when dealing with Big Data. In comparison with the state of the art, results demonstrate that DSVD-autoencoder provides a highly competitive solution to deal with very large data sets by reducing training from several hours to seconds while maintaining good accuracy.
The importance of understanding and explaining the associated classification results in the utilization of artificial intelligence (AI) in many different practical applications (e.g., cyber security and forensics) has...
详细信息
The importance of understanding and explaining the associated classification results in the utilization of artificial intelligence (AI) in many different practical applications (e.g., cyber security and forensics) has contributed to the trend of moving away from black-box / opaque AI towards explainable AI (XAI). In this article, we propose the first interpretable autoencoder based on decision trees, which is designed to handle categorical data without the need to transform the data representation. Furthermore, our proposed interpretable autoencoder provides a natural explanation for experts in the application area. The experimental findings show that our proposed interpretable autoencoder is among the top-ranked anomaly detection algorithms, along with one-class Support Vector Machine (SVM) and Gaussian Mixture. More specifically, our proposal is on average 2% below the best Area Under the Curve (AUC) result and 3% over the other Average Precision scores, in comparison to One-class SVM, Isolation Forest, Local Outlier Factor, Elliptic Envelope, Gaussian Mixture Model, and eForest.
A deep autoencoder (DAE)-based structure for end-to-end communication over the two-user Z-interference channel (ZIC) with finite-alphabet inputs is designed in this paper. The proposed structure jointly optimizes the ...
详细信息
A deep autoencoder (DAE)-based structure for end-to-end communication over the two-user Z-interference channel (ZIC) with finite-alphabet inputs is designed in this paper. The proposed structure jointly optimizes the two encoder/decoder pairs and generates interference-aware constellations that dynamically adapt their shape based on interference intensity to minimize the bit error rate (BER). An in-phase/quadrature-phase (I/Q) power allocation layer is introduced in the DAE to guarantee an average power constraint and enable the architecture to generate constellations with nonuniform shapes. This brings further gain compared to standard uniform constellations such as quadrature amplitude modulation. The proposed structure is then extended to work with imperfect channel state information (CSI). The CSI imperfection due to both the estimation and quantization errors are examined. The performance of the DAE-ZIC is compared with two baseline methods, i.e., standard and rotated constellations. The proposed structure significantly enhances the performance of the ZIC both for the perfect and imperfect CSI. Simulation results show that the improvement is achieved in all interference regimes (weak, moderate, and strong) and consistently increases with the signal-to-noise ratio (SNR). For instance, more than an order of magnitude BER reduction is obtained with respect to the most competitive conventional method at weak interference when SNR > 15 dB$ and two bits per symbol are transmitted. The improvements reach about two orders of magnitude when quantization error exists, indicating that the DAE-ZIC is more robust to the interference compared to the conventional methods.
暂无评论