Training autoencoders is non-trivial. Convergence to the identity function or over fitting are common pitfalls. Population based algorithms like coevolutionary algorithms can provide diversity. To more robustly train ...
详细信息
ISBN:
(纸本)9798400704949
Training autoencoders is non-trivial. Convergence to the identity function or over fitting are common pitfalls. Population based algorithms like coevolutionary algorithms can provide diversity. To more robustly train autoencoders, we introduce a novel cooperative coevolutionary algorithm that exploits a spatial topology. We investigate the impact of algorithm parameters and design choices on the performance. On a simple tunable benchmark problem we observe that the performance can be improved over that of an conventionally trained autoencoder. However, the training convergence can be slow, despite the final model performance being competitive with a conventional autoencoder.
Breast cancer is one of the most common cancers in human's life that frequently turn up in women as well as rarely in men too. The majority of the time, the results of the biopsy are used to identify the malignanc...
详细信息
ISBN:
(纸本)9798331540661;9798331540678
Breast cancer is one of the most common cancers in human's life that frequently turn up in women as well as rarely in men too. The majority of the time, the results of the biopsy are used to identify the malignancy and aid in establishing its stage. Convolution neural networks (CNN) and other Deep Learning (DL) architectures are mainly utilized for image classification. This even works well for classifying images of breast cancer. There are currently various feature extraction mechanisms available. Additionally, the CNN is utilized for feature extraction and image classification. During deployment stage, the adaptors have trained in transforming the test image as well as its features for minimizing the domain shift has been measured through the Convolutional autoencoder (CAE) reconstruction loss. This research has concentrated in building a model is adapted with a single test that subjected at inference and the proposed model has adopted neural networks as an AE as Transfer Learning (TL) that performs an image analysis task such as segmentation and even set as an adopter for pre training the model. The AE used to train from the source dataset and perform as the adaptors that have been optimized at the testing stage using a single test subject for effective computation. Therefore, this study has used VGG16 and VGG19 with CNN to extract features from BreaKHis database that involves images of microscopic biopsy to benign as well as malignant breast tumors for performing analysis of the unsupervised images. The evaluated results showed that accuracy of CAE with CNN-VGG16 has high accuracy as 96.17% in training that indicates DL models have appropriate in detection and classification of the breast cancers precisely.
Securing Internet of Things (IoT) devices against threats is crucial due to their significant impact on cyber-physical systems. Traditional intrusion detection systems often fall short in protecting the vast and diver...
详细信息
ISBN:
(纸本)9798350358810;9798350358803
Securing Internet of Things (IoT) devices against threats is crucial due to their significant impact on cyber-physical systems. Traditional intrusion detection systems often fall short in protecting the vast and diverse array of IoT devices. One key limitation is their lack of an anomaly detection objective, which is essential for identifying sophisticated threats that do not match known patterns. To address this critical gap, we have introduced a unique approach that utilizes an objective-based anomaly detection model. Our model, integrating a Deep Support Vector Data Description (DSVDD) with a Contractive autoencoder (CAE), named DSVDD-CAE, enhances the relevance of latent representations for anomaly detection and thereby improves accuracy. This innovative combination has significantly outperformed popular anomaly detection algorithms like KMeans, OCSVM, and Isolation Forest. On the ToN-IoT dataset, our method achieved a precision of 98.77%, a recall of 99.74%, an F1-score of 99.25%, and an accuracy of 99.57%. Similarly, on the IoTID20 dataset, it reached a precision of 98.25%, a recall of 99.80%, an F1-score of 99.01%, and an accuracy of 99.64%. These results demonstrate that our model excels in accurately detecting both known and novel IoT attacks, thereby significantly advancing the field of IoT security and providing a more resilient cyber-physical ecosystem.
This paper presents an autoencoder network designed to detect the severity of bridge damage using vibration signals obtained from passing vehicles. While one signal from a single vehicle may only contain limited infor...
详细信息
This paper presents an autoencoder network designed to detect the severity of bridge damage using vibration signals obtained from passing vehicles. While one signal from a single vehicle may only contain limited information about the structures, a large dataset with hundreds or thousands of signals can provide a substantial amount of information. The network is composed of an encoder with a convolutional layer and a long short-term memory (LSTM) layer and a decoder constructed with a fully connected (FC) layer to regenerate the initial input. In our approach, the particular novelty that sets it apart is the fact that we highlight the utilization of solely acceleration signals from intact bridges for training data, as intentionally damaging bridges to obtain signals for damaged conditions is infeasible and unethical in real-world scenarios. To evaluate the results, a metric called the variance-weighted coefficient of determination (R-2) is used to measure the goodness of the reconstruction of input signals and determine the damage severity of the bridge. The testing results indicate a strong linear relationship between the damage level and R-2. It should be noted that the vibration signals were collected from closed-form equations and numerical models with different combinations of bridge damage levels and vehicle properties. However, with the rapid growth of smartphones and data transmission technologies, the work can be extended to large amounts of real bridge data using such devices.
Hyperspectral unmixing is a crucial step in hyperspectral image processing. Hyperspectral images in real scenes are saturated with spectral variability, and unmixing performance is limited. We propose the Spectral Var...
详细信息
ISBN:
(纸本)9798350360332;9798350360325
Hyperspectral unmixing is a crucial step in hyperspectral image processing. Hyperspectral images in real scenes are saturated with spectral variability, and unmixing performance is limited. We propose the Spectral Variability Attention Net (SVA-Net). We have separately designed a Complementary Feature Enhancement Module (CFE) and a Spectral Variability Attention Mechanism to capture both the original material features in the image and other easily overlooked features. In addition, we design the improved mixing model based on augmented linear mixing model (ALMM) to better cope with the effects of spectral variability. Experiments on real datasets demonstrate the effectiveness of our model.
For the processing of echo signals in defect detection, wavelet transform and principal component analysis are mostly used to extract features. However, the feature values obtained by these methods often lead to redun...
详细信息
ISBN:
(纸本)9798350350319;9798350350302
For the processing of echo signals in defect detection, wavelet transform and principal component analysis are mostly used to extract features. However, the feature values obtained by these methods often lead to redundancy, resulting in the waste of a lot of resources for defect identification. This paper, in the context of defects in thick-walled steel plates with rough surfaces, proposes a defect category recognition classification method based on an autoencoder-BP neural network. It uses signals from electromagnetic ultrasonic and pulsed eddy current composite detection as the neural network learning signals. Impedance analysis is used to more comprehensively reflect the characteristics of defects, thereby improving the accuracy of defect identification. autoencoder is selected to extract the geometric features of the composite detection signals, which can effectively extract useful features from the dataset. The feature dataset is then divided into training and testing sets. Simulation experiments show that the trained neural network model has achieved a classification accuracy of 90.8% in the testing.
This paper introduces a novel architecture for the video object segmentation (VOS) challenge to achieve greater label efficiency. Previous studies have primarily tackled this problem through either match-based or prop...
详细信息
ISBN:
(纸本)9798400709777
This paper introduces a novel architecture for the video object segmentation (VOS) challenge to achieve greater label efficiency. Previous studies have primarily tackled this problem through either match-based or propagate-based architectures, relying on fully annotated datasets. In contrast, we propose the spatiotemporal mask autoencoder (STMAE), a novel VOS architecture constructed using annotations solely from the first frame. Specifically, STMAE generates a precise mask by initially aggregating a coarse mask from previous frames based on visual correspondence provided by an image encoder and then reconstructing it. We further propose a one-shot training strategy to learn general object representations for VOS using only the first frame mask. This strategy incorporates a reconstruction loss that guides the network to reconstruct the first frame mask from the spatiotemporal aggregation. Finally, extensive experiments conducted on the DAVIS and YouTube-VOS datasets demonstrate that STMAE achieves remarkable performance while effectively addressing the labor-intensive annotation issue.
Community detection intends to cluster graph nodes with relevant information, and community detection for attributed graphs is of great practical importance. However, the existing work is still insufficient in terms o...
详细信息
ISBN:
(纸本)9798350359329;9798350359312
Community detection intends to cluster graph nodes with relevant information, and community detection for attributed graphs is of great practical importance. However, the existing work is still insufficient in terms of the representation of network topology and attribute information. In addition, the impact of noise and missing attributes on attributed graph representation learning has been rarely considered. To address the above issues, this paper proposes an innovative masked dual graph autoencoder (MDGAE) method for attributed graph community detection. The method consists of a graph attention autoencoder module, a masked graph autoencoder module, a graph representation fusion module, and a self-optimizing clustering module. Firstly, MDGAE fully explores graph topological and attribute information through the graph attention autoencoder and the masked graph autoencoder, respectively, and employs a masking strategy to solve the problem of noise and missing attributes efficiently;then, the low-dimensional representations obtained from the two autoencoders are weighted and fused for clustering;finally, the clustering results are iteratively optimized by joining with the self-optimizing clustering module. The experimental results on three benchmark datasets demonstrate the advantages of the MDGAE method over the existing algorithms.
These days, one of the major downsides of Generalized Frequency Division Multiplexing (GFDM) systems is a high peak-to-average power ratio (PAPR). In this research, we present a novel deep learning autoencoder-based m...
详细信息
These days, one of the major downsides of Generalized Frequency Division Multiplexing (GFDM) systems is a high peak-to-average power ratio (PAPR). In this research, we present a novel deep learning autoencoder-based method to lower the PAPR of GFDM. The PAPR-reducing network (PRNet), also known as the PAPR-reducing method, is based on the encoder-decoder neural network (autoencoder). In the PAPR-reducing network (PRNet), the bit error rate (BER) and the PAPR of the GFDM system are jointly minimised by adaptively determining the constellation mapping and damping of symbols on each subcarrier and sub-symbol.
WiFi human sensing is highly regarded for its low-cost and privacy advantages in recognizing human activities. However, its effectiveness is largely confined to controlled, single-user, line-of-sight settings, limited...
详细信息
ISBN:
(纸本)9798350344868;9798350344851
WiFi human sensing is highly regarded for its low-cost and privacy advantages in recognizing human activities. However, its effectiveness is largely confined to controlled, single-user, line-of-sight settings, limited by data collection complexities and the scarcity of labeled datasets. Traditional cross-modal methods, aimed at mitigating these limitations by enabling self-supervised learning without labeled data, struggle to extract meaningful features from amplitude-phase combinations. In response, we introduce AutoSen, an innovative automatic WiFi sensing solution that departs from conventional approaches. AutoSen establishes a direct link between amplitude and phase through automated cross-modal autoencoder learning. This autoencoder efficiently extracts valuable features from unlabeled CSI data, encompassing amplitude and phase information while eliminating their respective unique noises. These features are then leveraged for specific tasks using few-shot learning techniques. AutoSen's performance is rigorously evaluated on a publicly accessible benchmark dataset, demonstrating its exceptional capabilities in automatic WiFi sensing through the extraction of comprehensive cross-modal features.
暂无评论