Variational Autoencoder Network Research Articles

Online nonlinear data reconciliation to enhance nonlinear dynamic process monitoring using conditional dynamic variational autoencoder networks with particle filters

In the chemical plants, data-driven process monitoring serves as a vital tool to ensure product quality and maintain production line safety. However, the accuracy of monitoring hinges directly upon the quality of process data. Given the inherently slow and complex nature of chemical processes, coupled with the potential for gross errors in process data leading to inaccuracies in model predictions, this paper proposes a method called Conditional Dynamic Variational Autoencoder combined with a Particle Filter (CDVAE-PF) for data reconciliation and subsequent process monitoring. CDVAE-PF leverages the capabilities of Conditional Dynamic Variational Autoencoder (CDVAE) to effectively model chemical process data in the presence of noise. This probabilistic model serves as the foundation for the Particle Filter (PF), which is employed for data reconciliation. Moreover, CDVAE-PF incorporates mechanisms to detect and rectify gross errors in process data, further enhancing its efficacy in data reconciliation. Subsequently, monitoring indices based on CDVAE are established to facilitate process monitoring. Through numerical simulations of a two-to-one variables Continuous Stirred Tank Reactor (CSTR) example and a fifteen-to-one variables dichloroethane distillation process from an actual chemical plant, CDVAE-PF demonstrates its effectiveness by reducing mean absolute error to 7.8 % and 12.8 % respectively in gross error data reconciliation. Moreover, in terms of monitoring performance, CDVAE-PF successfully mitigates misjudgments caused by gross errors, thereby significantly enhancing the reliability of process monitoring in chemical plants.

Image forgery detection by combining Visual Transformer with Variational Autoencoder Network

Recently, the applications and artificial intelligences used for image manipulation have become quite successful. In this case, the manipulation of personal data can lead to problems of insurmountable magnitude. Such problems not only put personal data at risk, but also lead us to unethical practices, with potentially irreversible negative consequences. For this reason, the reliability of image or video data is highly questionable. To solve this challenging problem, we introduce a Visual Transformer based Visual Transformer with Variational Autoencoder Network (ViT-VAE Net) model. The model includes Visual Transformer, one of the state-of-the-art architectures. In addition to this architecture, a Variational Auto Encoder structure is also included. is much more effective than models developed with the classical Convolutional Neural Network (CNN). Unlike models developed with CNN, it can perform operations on images of any size without being bound by a standard image resolution. In addition, thanks to the self-attention mechanism in the Visual Transformer architecture, manipulations on the image are caught more easily than CNN. The ViT-VAE Net model was trained with a large dataset and tested with 4 different datasets. With a success rate of 67 % on the training dataset, the model provided promising results. Very high rates were also obtained with the test datasets.

Melting phenomena of self-organized magnetic structures investigated by variational autoencoder

The phase transition phenomenon is an important research topic in various physical studies. However, it is difficult to define the order parameters in many complex systems involving self-organized structures. We propose a method to define order parameters using a variational autoencoder network. To demonstrate these capabilities, we trained a deep learning network with a dataset composed of spin configurations in a chiral magnetic system at various temperatures. It removes thermal fluctuations from the input data and leaves the remaining structural information with a spin magnitude. We define an order parameter with magnitude of output spins and compare the results with those of conventional analysis. The comparison indicates similar results. Using the order parameter, the thermal properties of the chiral magnetic system were investigated by varying the physical parameters and data size.

VAE-GAN3D: Leveraging image-based semantics for 3D zero-shot recognition

The current state of 3D zero-shot recognition falls short in performance when compared to its counterpart in 2D images. A major challenge is the absence of a robust feature extractor for 3D point cloud data. To overcome this challenge, a large dataset needs to be collected, processed, and fed into a deep-learning model capable of generating distinguishable features after training. To this end, we propose VAE-GAN3D, a model that uses a combination of a variational autoencoder (VAE) and a generative adversarial network (GAN) to supplement the small amount of data available for training with synthetic examples of classes. This allows for the underlying data distribution of large and complex data to be learned by the deep VAE network. When combined with a GAN, this network generates synthetic features that exhibit consistency in both seen and unseen classes. Furthermore, we notice that for tasks involving a small domain of classes, the existing text features do not contribute significantly to zero-shot learning. Therefore, we introduce image representation-based semantic features of classes, which improve the performance of zero-shot recognition for 3D objects. We assess the proposed model using three different datasets and present a technique for partitioning the RGB-D object dataset, which contains real-world objects, into seen and unseen classes. Our approach shows promising results in addressing the challenges of 3D zero-shot recognition and presents a novel solution for improving the accuracy of 3D point cloud recognition.

Structural Damage Detection Based on the Correlation of Variational Autoencoder Neural Networks Using Limited Sensors.

Identifying the structural state without baseline data is an important engineering problem in the field of structural health monitoring, which is crucial for assessing the safety condition of structures. In the context of limited accelerometers available, this paper proposes a correlation-based damage identification method using Variational Autoencoder neural networks. The approach involves initially constructing a Variational Autoencoder network model for bridge damage detection, optimizing parameters such as loss functions and learning rates for the model, and ultimately utilizing response data from limited sensors for model training analysis to determine the structural state. The contribution of this paper lies in the ability to identify structural damage without baseline data using response data from a small number of sensors, reducing sensor costs and enhancing practical applications in engineering. The effectiveness of the proposed method is demonstrated through numerical simulations and experimental structures. The results show that the method can identify the location of damage under different damage conditions, exhibits strong robustness in detecting multiple damages, and further enhances the accuracy of identifying bridge structures.

Relevance variable selection variational auto-encoder network for quality-related nonlinear process monitoring

Quality-related process monitoring is essential for revealing changes in product quality and ensuring industrial safety. Therefore, it is crucial to distinguish enough quality-related features within the data. To learn the nonlinear characteristics and obtain quality-related features in process data, a novel process monitoring method named relevance variable selection variational auto-encoder (RVS-VAE) is proposed. Firstly, to enhance the strong quality-related variables and learn the time correlation, the normalized original data is weighted by mutual information and augmented into data matrices. Secondly, the RVS strategy is proposed to select the most quality-related variables from the latent features. These quality-related features are further utilized to monitor the process. What is more, when a fault is detected, relevance variable relative contribution (RVRC) method is presented to locate the fault variables. Finally, performance of the proposed RVS-VAE method is evaluated through comparative experiments on tennessee eastman (TE) process.

A comparative study of variational autoencoders, normalizing flows, and score-based diffusion models for electrical impedance tomography

Abstract Electrical Impedance Tomography (EIT) is a widely employed imaging technique in industrial inspection, geophysical prospecting, and medical imaging. However, the inherent nonlinearity and ill-posedness of EIT image reconstruction present challenges for classical regularization techniques, such as the critical selection of regularization terms and the lack of prior knowledge. Deep generative models (DGMs) have been shown to play a crucial role in learning implicit regularizers and prior knowledge. This study aims to investigate the potential of three DGMs – variational autoencoder networks, normalizing flow, and score-based diffusion model – to learn implicit regularizers in learning-based EIT imaging. We first introduce background information on EIT imaging and its inverse problem formulation. Next, we propose three algorithms for performing EIT inverse problems based on corresponding DGMs. Finally, we present numerical and visual experiments, which reveal that (1) no single method consistently outperforms the others across all settings, and (2) when reconstructing an object with two anomalies using a well-trained model based on a training dataset containing four anomalies, the conditional normalizing flow (CNF) model exhibits the best generalization in low-level noise, while the conditional score-based diffusion model (CSD*) demonstrates the best generalization in high-level noise settings. We hope our preliminary efforts will encourage other researchers to assess their DGMs in EIT and other nonlinear inverse problems.

Deep Causality Variational Autoencoder Network for Identifying the Potential Biomarkers of Brain Disorders.

Identifying causality from observational time-series data is a key problem in dealing with complex dynamic systems. Inferring the direction of connection between brain regions (i.e., causality) has become the central topic in the domain of fMRI. The purpose of this study is to obtain causal graphs that characterize the causal relationship between brain regions based on time series data. To address this issue, we designed a novel model named deep causal variational autoencoder (CVAE) to estimate the causal relationship between brain regions. This network contains a causal layer that can estimate the causal relationship between different brain regions directly. Compared with previous approaches, our method relaxes many constraints on the structure of underlying causal graph. Our proposed method achieves excellent performance on both the Alzheimer's Disease Neuroimaging Initiative (ADNI) and the Autism Brain Imaging Data Exchange 1 (ABIDE1) databases. Moreover, the experimental results show that deep CVAE has promising applications in the field of brain disease identification.

Incorporating variational auto-encoder networks for text-driven generation of 3D motion human body

Incorporating variational auto-encoder networks for text-driven generation of 3D motion human body

APGVAE: Adaptive disentangled representation learning with the graph-based structure information

Neural networks are used to learn task-oriented high-level representations in an end-to-end manner by building a multi-layer neural network. Generation models have developed rapidly with the emergence of deep neural networks. But it still has problems with the insufficient authenticity of generated images, the deficiency of diversity, consistency, and unexplainability in the generation process. Disentangled representation is an effective method to learn a high-level feature representation and realize the interpretability of deep neural networks. We propose a general disentangled representation learning network with variational autoencoder network as the basic framework for the image generation process. The graph-based structure of the priors is embedded in the last module of the deep encoder network to build the feature spaces by the class, task-oriented, and task-unrelated information respectively. Meanwhile the priors should be adaptively modified with the task relevance of a generated image. And the semi-supervised learning is further involved in the disentangled representation network framework to reduce the requirements of label and extend the majority of feature space under the task-unrelated feature assumption. Experimental results show that the proposed method is efficient for various types of images and has a good potential for further research and development.

Flow Field Reconstruction of 2D Hypersonic Inlets Based on a Variational Autoencoder

The inlet is one of the most important components of a hypersonic vehicle. The design and optimization of the hypersonic inlet is of great significance to the research and development of hypersonic vehicles. In recent years, artificial intelligence techniques have been used to improve the efficiency of aerodynamic optimization. Deep generative models, such as variational autoencoder (VAE) and generative adversarial network (GAN), have been used in a variety of flow problems in the last two years, making fast reconstruction and prediction of the full flow field possible. In this study, a hybrid multilayer perceptron (MLP) combined with a VAE network is used to reconstruct and predict the flow field of a two-dimensional multiwedge hypersonic inlet. The obtained results show that the VAE network can reconstruct the overall flow structure of the hypersonic flow field with high accuracy. The reconstruction accuracy of complex flow structures, such as shockwaves, boundary layers, and separation bubbles, is satisfactory. The flow field prediction model based on the MLP-VAE hybrid model has a strong generalization and generation ability, achieving relatively accurate flow field prediction for inlets with geometric configurations outside the training set.

Anomalous Change Detection in Drilling Process Using Variational Autoencoder with Temperature Near Drill Edge

The different flexibility and diversity requirements for respective manufacturing units have made modern cutting tool management much more crucial and complicated, as a greater variety of tools and more frequent tool changes are required to enhance production efficiency and avoid unplanned manufacturing downtime. Developing in-process anomalous change detection methods has been identified as an essential challenge. Machine learning techniques have been widely applied in tool condition monitoring and anomalous change detection. As anomaly data is rare in manufacturing processes, supervised machine learning approaches (such as regression and classification) are not applied to the anomalous change detection problem. Rather, self-supervised machine learning (a representative type of unsupervised machine learning) is applied. This study describes a variational autoencoder (VAE) neural network and proposes a VAE-based method for tool condition monitoring and change detection in a drilling process using the temperature near a drill edge. The proposed VAE evaluates the drill tool condition based on the reconstruction error between the input temperature and its estimate per a drill unit process through the trained network. Computational simulations demonstrate that the proposed VAE network model can avoid overfitting to the anomaly data and that its expressive power is greater than that of the conventional autoencoder model.

Multiresolution Fusion Convolutional Network for Open Set Human Activity Recognition

In recent years, sensor-based human activity recognition (HAR) technology has been the focus of extensive research and has been successfully applied to many aspects of people’s lives, but there are still some deficiencies. Most studies only distinguish daily activities and have low accuracy for easy confusing activities. In addition, many deep learning models only consider closed set HAR, but the real world contains unknown class activities that cannot be foreseen, which makes it challenging to apply these models to practice. In view of the above problems, this paper proposes a multi-resolution fusion convolution network (MRFC-Net) to cover the shortcoming that confusing activities are difficult to correctly identify, thus improving the accuracy of recognition. Furthermore, a multi-resolution fusion convolution variational auto-encoder network (MRFC-VAE-Net) for open set HAR is proposed. According to the reconstruction loss of the network, the corresponding threshold is set to effectively classify the known and unknown class activities in the open set. At the same time, a rich data set named daily-abnormal activity of special groups (DAASG) is constructed, which can be applied to the daily monitoring of special groups such as prisoners and the elderly. Experiments and analyses are carried out on the wireless sensor data mining (WISDM), physical activity monitoring for aging people (PAMAP2) and DAASG data sets, to prove the effectiveness and superiority of the proposed networks.

The prediction of drug sensitivity by multi-omics fusion reveals the heterogeneity of drug response in pan-cancer

Cancer drug response prediction based on genomic information plays a crucial role in modern pharmacogenomics, enabling individualized therapy. Given the expensive and complexity of biological experiments, computational methods serve as effective tools in predicting cancer drug sensitivity. In this study, we proposed a novel method called Multi-Omics Integrated Collective Variational Autoencoders (MOICVAE), which leverages integrated omics knowledge, including genomic and transcriptomic data, to fill in missing cancer-drug associations and enhance drug sensitivity prediction. Our method employs an encoder-decoder network to learn latent feature representations from cell lines. These learned feature vectors are then fed into a collective variational autoencoder network to train an association matrix. We evaluated MOICVAE on the GDSC and CCLE benchmark datasets using 10-fold cross-validation and achieved impressive AUCs of 0.856 and 0.808, respectively, outperforming state-of-the-art methods. Furthermore, on the TCGA dataset, consisting of 25 drugs across 7 cancer types, MOICVAE exhibited an average AUC of 0.91 in predicting drug sensitivity. Additionally, significant differences were observed in survival, tumor inflammatory assessment, and tumor microenvironment between the predicted drug-sensitive and drug-resistant groups. These results are consistent with predictions made on the METABRIC dataset. Moreover, we discovered that fusing omics data based on mRNA and CNV (copy number variations) yielded superior results in drug sensitivity prediction. MOICVAE not only achieved higher accuracy in drug sensitivity prediction but also provided additional value for combining immunotherapy with chemotherapy, offering patients with more precise treatment options. The code and dataset for MOICVAE are freely available at https://github.com/wanggnoc/MOICVAE.

Functional Connectivity Networks with Latent Distributions for Mild Cognitive Impairment Identification.

This work presents a novel approach to estimate brain functional connectivity networks via generative learning. Due to the complexity and variability of rs-fMRI signal, we consider it as a random variable, and utilize variational autoencoder networks to encode it as a confidence distribution in the latent space rather than as a fixed vector, so as to establish the relationship between them. First, the mean time series of each brain region of interest is mapped into a multivariate Gaussian distribution. The correlation between two brain regions is measured by the Jensen-Shannon divergence that describes the statistical similarity between two probability distributions, and then the adjacency matrix is created to indicate the functional connectivity strength of pairwise brain regions. Meanwhile, our findings show that the adjacency matrices obtained at VAE latent spaces of different dimensionalities have good complementarity for MCI identification in precision and recall, and the classification performance can be further boosted by an efficient cascade of classifiers. This proposal constructs brain functional networks from a statistical modeling standpoint, improving the statistical ability of population data and the generalization ability of observation data variability. We evaluate the proposed framework over the task of identifying subjects with MCI from normal controls, and the experimental results on the public dataset show that our method significantly outperforms both the baseline and current state-of-the-art methods.

Degradation trend feature generation improved rotating machines RUL prognosis method with limited run-to-failure data

The success of rotating machines’ data-driven remaining useful life (RUL) prognosis approaches depends heavily on the abundance of entire life cycle data. However, it is not easy to obtain sufficient run-to-failure data in industrial practice. Data generation technology is a promising solution for enriching data but fails to address the intrinsic complexity of nonlinear stage degradation and the time correlation of long-term data. This research proposes an RUL prognosis approach improved by the degradation trend feature generation variational autoencoder. First, this study develops a framework combining degradation trend generation features to resolve the issue of capturing the elements of time distribution for run-to-failure data. Second, a generation variational autoencoder network with a tendency block is proposed to create high-quality time series data correlation features. Third, original and created degradation trend features are subjected to deep adaptive fusion and health indicator extraction. A bi-directional long short-term memory network is employed to predict the degradation trend and obtain the RUL prognosis. Finally, the proposed approach’s feasibility is confirmed by cross-validation experiments on a bearing dataset, which reduces the prediction error by 22.309%.

Facial expression generation based on variational AutoEncoder network and cloud computing

With the rapid development of portable terminal devices, such as smartphones, the development of expression generation technology makes the emotional communication between people in social networks more rich and diverse. However, there are still great challenges in deploying deep networks on mobile phones with limited resources. Moreover, the muscle movement of the face produces different facial expressions. To this end, this paper proposes an efficient emotion image generation framework based on Transformer and variational auto‐encoders (VAE) on the cloud platform. Specifically, we collect human face which is further transferred to the service. To exploit connections between muscles and expressions, we introduce the multi‐head attention mechanism to construct Transformer. To generate different emotional images, we design a novel emotion generation model named Transformer‐embedded VAE (TEVAE). All the experimental results show that the proposed TEVAE obtains higher performance in emotion image generation tasks compared to VAE.

Graph embedding and Gaussian mixture variational autoencoder network for end-to-end analysis of single-cell RNA sequencing data

Single-cell RNA sequencing (scRNA-seq) is a revolutionary technology to determine the precise gene expression of individual cells and identify cell heterogeneity and subpopulations. However, technical limitations of scRNA-seq lead to heterogeneous and sparse data. Here, we present autoCell, a deep-learning approach for scRNA-seq dropout imputation and feature extraction. autoCell is a variational autoencoding network that combines graph embedding and a probabilistic depth Gaussian mixture model to infer the distribution of high-dimensional, sparse scRNA-seq data. We validate autoCell on simulated datasets and biologically relevant scRNA-seq. We show that interpolation of autoCell improves the performance of existing tools in identifying cell developmental trajectories of human preimplantation embryos. We identify disease-associated astrocytes (DAAs) and reconstruct DAA-specific molecular networks and ligand-receptor interactions involved in cell-cell communications using Alzheimer's disease as a prototypical example. autoCell provides a toolbox for end-to-end analysis of scRNA-seq data, including visualization, clustering, imputation, and disease-specific gene network identification.

Exploiting the sparse characteristics in probabilistic feature space for hyperspectral anomaly detection

Nowadays, the low-rank representation (LRR) and deep learning-based methods have received much attention in anomaly detection for hyperspectral images (HSIs). However, most of these methods mainly focus on the powerful reconstruction capability of the neural networks while ignoring the potential probability distribution of both anomalies and background pixels. To solve the problem, we propose a sparse component extraction-based probability distribution representation detector (SC-PDRD) framework, which integrates the characteristic of the sparse component obtained by the LRR model with the powerful probability representation ability of the variational autoencoder (VAE) network. The LRR model effectively separates the anomaly component from the background, which also serves as the prior anomaly distribution for each pixel. Moreover, the VAE architecture tries to recover the potential anomaly distribution using the sparse detection map in the feature space. In addition, we employ the Chebyshev neighborhood to leverage spatial information. The modified Wasserstein distance measures the distance between the test pixel and its neighborhood. The final detection map is attained by combining the prior anomalous degree of the anomalies with the output of the VAE network. Experimental results on three real HSIs demonstrate the effectiveness and superiority of SC-PDRD.

Linking properties to microstructure in liquid metal embedded elastomers via machine learning

Liquid metals (LM) are embedded in an elastomer matrix to obtain soft composites with unique thermal, dielectric, and mechanical properties. They have applications in soft robotics, biomedical engineering, and wearable electronics. By linking the structure to the properties of these materials, it is possible to perform material design rationally. Liquid-metal embedded elastomers (LMEEs) have been designed for targeted electro-thermo-mechanical properties by semi-supervised learning of structure–property (SP) links in a variational autoencoder network (VAE). The design parameters are the microstructural descriptors that are physically meaningful and have affine relationships with the synthetization of the studied particulate composite. The machine learning (ML) model is trained on a generated dataset of microstructural descriptors with their multifunctional property quantities as their labels. Sobol sequence is used for in-silico Design of Experiment (DoE) by sampling the design space to generate a comprehensive dataset of 3D microstructure realizations via a packing algorithm. The mechanical responses of the generated microstructures are simulated using a previously developed Finite Element (FE) model, considering the surface tension induced by LM inclusions, while the linear thermal and dielectric constants are homogenized with the help of our in-houseFast Fourier Transform (FFT) package. Following the training by minimization of an appropriate loss function, the VAE encoder acts as the surrogate of numerical solvers of the multifunctional homogenizations, and its decoder is used for the material design. Our results indicate the satisfactory performance of the surrogate model and the inverse calculator with respect to high-fidelity numerical simulations validated with LMEE experimental results.