Neural Network-based Identification Research Articles

Neural Network-Based Design of Two-Focal-Spot Terahertz Diffractive Optical Elements

Neural Network-Based Design of Two-Focal-Spot Terahertz Diffractive Optical Elements

Graph neural network-based identification of ditch matching patterns across multi-scale geospatial data

Ditches are vital to water system data. To ease the task of matching multi-scale ditch data and enhance accuracy (Acc), it is essential to discern ditch data matching patterns. Despite its importance, limited research has been conducted on ditch matching patterns, and the primary reasons for variations in multi-scale geographical entities’ matching patterns remain underexplored. Here, we introduce a supervised graph neural network (GNN) method, targeting the direct analysis of 1:10,000 ditch characteristics to identify the matching patterns between 1:10,000 and 1:25,000 datasets. The ditch network is depicted as a graph structure, with nodes representing ditch segments and edges indicating their connections. Subsequently, each ditch segment’s geometric, semantic, and topological attributes are computed as node attributes, and their matching patterns with the 1:25,000 dataset are labeled as node annotations. Ditch segment matching patterns are then categorized using supervised learning. Experiments using Overijssel, Netherlands’ ditch data reveal that this method achieves a 97.3% classification Acc, outperforming other GNN methods by 25.6–26.3% and traditional machine learning methods by 33%. These findings underscore the efficacy and superiority of the proposed supervised GNN approach in pinpointing ditch matching patterns.

A recurrent neural network-based identification of complex nonlinear dynamical systems: a novel structure, stability analysis and a comparative study

A recurrent neural network-based identification of complex nonlinear dynamical systems: a novel structure, stability analysis and a comparative study

Intelligent identification method of chemical processes based on maximum mean discrepancy domain generalization

BackgroundBenefiting from the massive process data stored in the plant distributed control system, the identification of chemical processes using deep learning methods has been widely studied. However, the statistical characteristics of chemical process data often change over time due to random interference caused by environmental impacts, the complexity and variability of chemical reaction processes, and changes in operating conditions. This makes the model obtained by the deep learning method based on the assumption of independent identical distribution sensitive to distribution changes. MethodsA neural network identification method based on multi-domain multi-kernel maximum mean discrepancy (MD-MK-MMD) domain generalization is proposed. The identification data is split with the maximum average distance to maximize the difference in the distribution of identification data. The MD-MK-MMD domain generalization method is adopted to reduce distribution difference and learn model parameters. Based on the gate recurrent unit (GRU) model and combined with MD-MK-MMD domain generalization, an intelligent identification framework for chemical processes is constructed. Significant findingsExperiments on actual process datasets of a heating furnace and a fractionator show the effectiveness of the proposed chemical process identification framework. The errors of the MD-MK-MMD domain generalization based identification method on the test sets are lower than that of the state-of-the-art neural network-based identification methods, which means the proposed method reduces the model sensitivity to the identification data distribution changes.

Graph Neural Network-Based Design Decision Support for Shared Mobility Systems

Abstract Emerging shared mobility systems are gaining popularity due to their significant economic and environmental benefits. In this paper, we present a network-based approach for predicting travel demand between stations (e.g., whether two stations have sufficient trips to form a strong connection) in shared mobility systems to support system design decisions. In particular, we answer the research question of whether local network information (e.g., the network neighboring station’s features of a station and its surrounding points of interest (POI), such as banks, schools, etc.) would influence the formation of a strong connection or not. If so, to what extent do such factors play a role? To answer this question, we propose using graph neural networks (GNNs), in which the concept of network embedding can capture and quantify the effect of local network structures. We compare the results with a regular artificial neural network (ANN) model that is agnostic to neighborhood information. This study is demonstrated using a real-world bike sharing system, the Divvy Bike in Chicago. We observe that the GNN prediction gains up to 8% higher performance than the ANN model. Our findings show that local network information is vital in the structure of a sharing mobility network, and the results generalize even when the network structure and density change significantly. With the GNN model, we show how it supports two crucial design decisions in bike sharing systems, i.e., where new stations should be added and how much capacity a station should have.

Designing Theoretical Shipborne ADCP Survey Trajectories for High-Frequency Radar Based on a Machine Learning Neural Network

A machine learning neural network-based design for shipborne ADCP navigation is proposed to improve the quality of high-frequency radar measurements. In traditional inversion algorithms for HF radars, sea surface velocity is directly extracted from electromagnetic echoes without constraints from oceanographic processes. Hence, we incorporated oceanographic information from observational data into seabed radar inversion results via an LSTM neural network model to enhance data accuracy. Through a series of numerical simulation experiments, we showed improved data accuracy and feasibility by incorporating both fixed-point and navigation observational data. The results indicate a significant reduction in (related) errors. This study has implications for guiding future navigation observations.

Automatic Gripper-Finger Design, Production, and Application: Toward Fast and Cost-Effective Small-Batch Production

Tactile robot-based assembly line adaptation of new products is still limited by the manual redesign, manufacturing, and exchange of the end-effector setup since the gripper fingers must often be adapted to the geometry of the product components to ensure a successful assembly process. In this article we present an automatic finger design, production, and evaluation pipeline, developed to improve this adaptation process. Two different form-closure–based design principles have been implemented to automatically generate the fingertip geometry: a projected surface representation-based approach as well as a Bézier surface fitting strategy. The resulting fingertips are printed via an automatic production unit and experimentally evaluated based on pick and insertion tasks for three different manipulation objects. To illustrate the potential usage of the introduced design methods for machine learning-based fingertip design approaches, we set up the training and testing process for a neural network-based design method. The proposed automatic fingertip design, production, and application framework represent a step further toward small-batch–size production, since the assembly adaptation effort, flexibility, and scalability are significantly improved.

A Problem-tailored Adversarial Deep Neural Network-Based Attack Model for Feed-Forward Physical Unclonable Functions

With the exceeding advancement in technology, the sophistication of attacks is considerably increasing. Standard security methods fall short of achieving the security essentials of IoT against physical attacks due to the nature of IoTs being resource-constrained elements. Physical Unclonable Functions (PUFs) have been successfully employed as a lightweight memoryless solution to secure IoT devices. PUF is a device that exploits the integrated circuits’ inherent randomness originated during the fabrication process to give each physical entity a unique identifier. Nevertheless, because PUFs are vulnerable to mathematical clonability, Feed-Forward Arbiter PUF (FF PUF) was introduced to withstand potential attack methods. Motivated by the necessity to expose a critical vulnerability of the standard FF PUFs design, we introduce a problem-tailored adversarial model to attack FF PUF design using a carefully engineered loop-specific neural network-based design calibrated and trained using FPGA-based in-silicon implementation data to exhibit real-world attacking scenarios posed on FF PUFs, in addition to applying simulated data. The empirical results show that the proposed adversarial model adds outperforming results to the existing studies in attacking FF PUFs, manifesting the improved efficiency in breaking FF PUFs. We demonstrate our high-performing results in numerical experiments of language modeling using the deep Neural Networks method.

Automated deep neural network-based identification, localization, and tracking of cardiac structures for ultrasound-guided interventional surgery.

The increase in the use of ultrasound-guided interventional therapy for cardiovascular diseases has increased the importance of intraoperative real-time cardiac ultrasound image interpretation. We thus aimed to develop a deep learning-based model to accurately identify, localize, and track the critical cardiac structures and lesions (9 kinds in total) and to validate the algorithm's performance using independent data sets. This diagnostic study developed a deep learning-based model using data collected from Fuwai Hospital between January 2018 and June 2019. The model was validated with independent French and American data sets. In total, 17,114 cardiac structures and lesions were used to develop the algorithm. The model findings were compared with those of 15 specialized physicians in multiple centers. For external validation, 516,805 tags and 27,938 tags were used from 2 different data sets. Regarding structure identification, the area under the receiver operating characteristic curve (AUC) of each structure in the training data set, optimal performance in the test data set, and median AUC of each structure identification were 1 (95% CI: 1-1), 1 (95% CI: 1-1), and 1 (95% CI: 1-1), respectively. Regarding structure localization, the optimal average accuracy was 0.83. As for structure identification, the accuracy of the model significantly outperformed the median performance of the experts (P<0.01). The optimal identification accuracies of the model in 2 independent external data sets were 89.5% and 90%, respectively (P=0.626). The model outperformed most human experts and was comparable to the optimal performance of all human experts in cardiac structure identification and localization, and could be used in the external data sets.

Neural Network-Based Identification of Cloud Types from Ground-Based Images of Cloud Layers

Clouds are a significant factor in regional climates and play a crucial role in regulating the Earth’s water cycle through the interaction of sunlight and wind. Meteorological agencies around the world must regularly observe and record cloud data. Unfortunately, the current methods for collecting cloud data mainly rely on manual observation. This paper presents a novel approach to identifying ground-based cloud images to aid in the collection of cloud data. However, there is currently no publicly available dataset that is suitable for this research. To solve this, we built a dataset of surface-shot images of clouds called the SSC, which was overseen by the Macao Meteorological Society. Compared to previous datasets, the SSC dataset offers a more balanced distribution of data samples across various cloud genera and provides a more precise classification of cloud genera. This paper presents a method for identifying cloud genera based on cloud texture, using convolutional neural networks. To extract cloud texture effectively, we apply Gamma Correction to the images. The experiments were conducted on the SSC dataset. The results show that the proposed model performs well in identifying 10 cloud genera, achieving an accuracy rate of 80% for the top three possibilities.

Metro Track Geometry Defect Identification Model Based on Car-Body Vibration Data and Differentiable Architecture Search

Efficient and low-cost modes for detecting metro track geometry defects (TGDs) are essential for condition-prediction-based preventive maintenance, which can help improve the safety of metro operations and reduce the maintenance cost of metro tracks. Compared with the traditional TGD detection method that utilizes the track geometry car, the method that uses a portable detector to acquire the car-body vibration data (CVD) can be used on an ordinary in-service train without occupying the metro schedule line, thereby improving efficiency and reducing the cost. A convolutional neural network-based identification model for TGD, built on a differentiable architecture search, is proposed in this study to employ only the CVD acquired by a portable detector for integrated identification of the type and severity level of TGDs. Second, the random oversampling method is introduced, and a strategy for applying this method is proposed to improve the poor training effect of the model caused by the natural class-imbalance problem arising from the TGD dataset. Subsequently, a comprehensive performance-evaluation metric (track geometry defect F-score) is designed by considering the actual management needs of the metro infrastructure. Finally, a case study is conducted using actual field data collected from Beijing Subway to validate the proposed model.

Memory Recurrent Elman Neural Network-Based Identification of Time-Delayed Nonlinear Dynamical System

In this article, an attempt has been made to propose an improved version of the classical Elman neural network (ENN) and its application is presented to identify the unknown dynamics of time-delayed nonlinear plants. The model proposed, known as memory recurrent ENN (MRENN), consists of an additional number of weighted self-feedback loops, an extra output context layer, and weighted connections of input signals (through adjustable weights) to the output-layer neuron. The model is proposed to be given only two signals as its inputs: plant unit-delayed value and the present value of the externally applied signal irrespective of the actual order of the plant (which most of the time may not be known). To guarantee stability, the parameters of the MRENN model are updated using the equations that are obtained by applying Lyapunov-stability criteria. Furthermore, the recursive learning rate scheme is constructed for speeding up the learning process. From the simulation results, MRENN model appears to have produced comparably superior outcomes when the performance of the suggested model is compared with that of other well-known models.

A graph based deep learning technology application in degenerative polyarthritis associated genes prediction

Abstract: Degenerative polyarthritis is the most common joint disease and affects millions of people worldwide. However, there is currently no cure for degenerative polyarthritis and no effective methods to prevent or slow down its progression. Gene regulatory relationships are vital for understanding disease mechanisms and developing treatment and novel drugs. Gene regulatory networks can be obtained from the RNA sequencing. Although various single-cell and bulk RNA sequencing data are available, an effective method to integrate the data for molecular diagnosis and treatment of degenerative polyarthritis has not yet been carried out. Here, we propose a novel deep learning-based method to efficiently capture the gene regulatory features of degenerative polyarthritis. First, we integrate single-cell RNA sequencing data-based gene regulatory network to model the gene regulatory relationships between genes and transcription factors as node feature aggregation. Second, we propose a graph convolutional model named dpTF-GCN on gene regulatory graph to transmit and update the node feature for potential associated genes predicting. According to the results, dpTF-GCN achieved the best performance among represented network-based methods. Furthermore, case studies suggest that dpTF-GCN can identify potential associated genes accurately. Our research not only provides theoretical and methodological support for the study of degenerative polyarthritis, but also provides a research case for the application of graph neural network-based identification of associated genes in other diseases.

Variational autoencoder-based topological optimization of an anechoic coating: An efficient- and neural network-based design

An anechoic coating is an artificial heterogeneous composite material composed of periodic cells with cavities. Using the local resonance of cavities to reduce their sound absorption frequencies and widen their frequency bands has been a research hotspot in recent years. One of the main challenges involves optimizing the cavity structure of an anechoic coating to obtain low-frequency, broadband, and strong sound absorption properties. In this paper, a variational autoencoder (VAE) model-based topology optimization method was investigated. First, the finite-element method (FEM) was used to calculate the sound absorption coefficient, and a dataset was constructed with samples whose average sound absorption coefficients ranged from 200 to 6000 Hz and were greater than 0.75. Then, the VAE model was trained to learn the key features of an anechoic coating. Finally, the data were reconstructed with Gaussian distributions. The decoder network of the trained VAE model was used to design a new anechoic coating. It took only approximately 3 s to generate 100 new topologies, and the average absorption coefficients were all greater than 0.754. This efficient neural network-based method can be further generalized to optimize the designs of various mechanical structural materials with specific functions.

Artificial Neural Network-Based Identification of Associations between UCP2 and UCP3 Gene Polymorphisms and Meat Quantity Traits

In identifying mutations occurring in distinct cow breeds, genetic elements must be taken into consideration. More recently, these hereditary features have gained attention throughout the world. As in many underdeveloped nations, to bridge the deficit in molecular genetics, multiple solutions are required. The inner membrane anion carrier superfamily contains the uncoupling proteins (UCPs), vital to energy regulation. Research on heredity has shown that variations in the UCP2 and UCP3 genes are connected to obesity and metabolic syndrome. This research aimed to investigate if any mutation in the UCP 2 and UCP 3 genes are related to many characteristics in Pakistan’s three indigenous cattle breeds using artificial neural network (ANN). For better analysis, the output of the ANN model is loaded into the Primer Premier 3 software. Using polymerase chain reaction-single strand conformation polymorphism (PCR-SSCP) and sequencing, the results of this study indicated 07 variations in the exon 4 region of the UCP2 gene and 03 variants in the exon 3 area of the UCP3 gene among 215 indigenous cow breeds. The association study revealed that the g.C35G mutation in the UCP3 gene is strongly related to meat quantity characteristics such as carcass weight and drip percentage (P0.05) but not with body height or hip width ( P &gt; 0.05 ). Sequence analysis showed five distinct diplotypes: AA, BC, AC, CC, and CD. Cattle with the novel heterozygous diplotype BC perform better in carcass trait and drip percentage than animals with other genotypes. The study’s findings suggest that the UCP3 gene may be utilized for marker-assisted selection (MAS) and breed mixing in Pakistan cattle breeds to aid in the country’s economic growth.

Differential Neural Network-Based Nonparametric Identification of Eye Response to Enforced Head Motion

Dynamic motion simulators cannot provide the same stimulation of sensory systems as real motion. Hence, only a subset of human senses should be targeted. For simulators providing vestibular stimulus, an automatic bodily function of vestibular–ocular reflex (VOR) can objectively measure how accurate motion simulation is. This requires a model of ocular response to enforced accelerations, an attempt to create which is shown in this paper. The proposed model corresponds to a single-layer spiking differential neural network with its activation functions are based on the dynamic Izhikevich model of neuron dynamics. An experiment is proposed to collect training data corresponding to controlled accelerated motions that produce VOR, measured using an eye-tracking system. The effectiveness of the proposed identification is demonstrated by comparing its performance with a traditional sigmoidal identifier. The proposed model based on dynamic representations of activation functions produces a more accurate approximation of foveal motion as the estimation of mean square error confirms.

Deep neural network-based identification of driving risk utilizing driver dependent vehicle driving features: a scheme for critical infrastructure protection

Deep neural network-based identification of driving risk utilizing driver dependent vehicle driving features: a scheme for critical infrastructure protection

Acoustic topology optimization using moving morphable components in neural network-based design

Acoustic topology optimization using moving morphable components in neural network-based design

Neural Network-Based Identification of a PSA Process for Production and Purification of Bioethanol

The pressure swing adsorption (PSA) process, is a novel method for the purification and production of bioethanol. A highly non-linear rigorous model is implemented to simulate the cyclic dynamics of PSA, achieving purity of 99&#x0025; wt of ethanol, which meets international standards to be used as fuel. The contribution of this work focuses on obtaining an identified model capable of capturing the important dynamics of the PSA process (with a fit above of 90&#x0025;) and to be used for controller design purposes, since it is very complicated to design control in highly nonlinear models that are represented with partial differential equations (PDE). For proof of concept, a comparison between Hammersetein-Wienner and Artificial Neural-Networks showed the relevance of the proposed method, using the same input and output signals. Both identified models capture the important dynamics of the rigorous PSA model.

A hybrid artificial neural network-based identification systemfor fine-grained composites

Recent interest in the development of innovative building materials has brought about the need for a detailed assessment of their mechanical fracture properties. The parameters for these need to be acquired, and one of the possible ways of doing so is to obtain them indirectly - based on a combination of fracture testing and inverse analysis. The paper describes a method for the identification of selected parameters of mortars and other fine-grained brittle matrix composites. The cornerstone of the method is the use of an artificial neural network, which is utilized as a surrogate model of the inverse relation between the measured specimen response parameters and the sought material parameters. Due to the potentially wide range of composite mixtures and hence the wide range of experimental responses likely to be gained from individual specimens, an ensemble of artificial neural networks was created. It allows the entire range of variants to be covered and provides resulting parameter values with sufficient precision. Such a system is also easy to expand if a composite with properties outside the current range is tested. The capabilities of the proposed identification system are demonstrated on two selected types of fine-grained composites with different specimen responses. The first group of specimens was made of composite based on alkali-activated slag with standardized and natural sand investigated within the time interval of 3 to 330 days of aging. The second tested composite contained alkali-activated fly ash matrix, and the effect of the addition of natural fibers on fracture response was investigated.