Neural Model Research Articles

Non-contact monitoring of human cardiorespiratory activity during sleep using FMCW millimeter wave radar

Millimeter wave radar can be utilized for non-contact, long-term, and accurate health monitoring. This paper presents an application for vital signs detection including respiratory rate (RR) and heart rate (HR) during sleep, as well as facilitating rapid detection and extended recording of apnea syndrome. This paper employs frequency spectrum analysis for RR estimation and develops an algorithm of template matching estimation (TME) for real-time, accurate HR estimation by comparing radar signal fragments to a template with a frequency signature. The time required for HR estimation is reduced to 1 s. In addition, a neural network-based model was developed for detecting the apnea events with the accuracy of 93.60% and calculates the duration in seconds. The system underwent testing using irregular analog signals of vital signs. In the actual scene testing, the accuracy of the RR estimation is 90.58%, and 90.25% for HR estimation.

Interpretable neural network classification model using first-order logic rules

Over the past decade, the field of neural networks has made significant strides, particularly in deep learning. However, their limited interpretability has constrained their application in certain critical domains, drawing widespread criticism. Researchers have proposed various methods for explaining neural networks to address this challenge. This paper focuses on rule-based explanations for neural network classification problems. We propose IRCnet, a scalable classification model based on first-order logic rules. IRCnet consists of layers for learning conjunction and disjunction rules, utilizing binary logic activation functions to enhance interpretability. The model is initially trained using a continuous-weight version, which is later binarized to produce a discrete-weight version. During training, we innovatively employed gradient approximation method to handle the non-differentiable weight binarization function, thereby enabling the training of split matrices used for binarization. Finally, Conjunctive Normal Form (CNF) or Disjunctive Normal Form (DNF) rules are extracted from the model’s discrete-weight version. Experimental results indicate that our model achieves the highest or near-highest performance across various classification metrics in multiple structured datasets while demonstrating significant scalability. It effectively balances classification accuracy with the complexity of the generated rules.

A novel approach for tool-narayanaswamy-moynihan model parameter extraction using multi-scale neural model

The accurate determination of parameters in the Tool-Narayanaswamy-Moynihan (TNM) model, which describes the viscoelastic behavior of glass-forming materials, is crucial for predicting material responses through various thermal histories.Traditional methods rely heavily on curve-fitting techniques; however, these often fail due to noise in the data. Furthermore, traditional methods are computationally intensive and prone to inaccuracies, particularly when dealing with complex datasets or when the initial parameter guesses are far from optimal; also, they require a skilled personnel.In this study, we propose the application of a multi-scale convolutional neural network (MCNN) as a machine learning approach to address these challenges. The MCNN model is trained on a comprehensive simulated dataset encompassing a wide range of TNM parameters, allowing it to learn intricate patterns and dependencies within the data that are difficult to capture with conventional methods. Our results show that the MCNN significantly improves the accuracy of the parameter estimations for β and x across the entire spectrum of tested conditions, achieving performance that is not only comparable to, but often surpasses, traditional curve-fitting methods. Furthermore, the MCNN demonstrates superior robustness when initial parameter estimates are suboptimal or when the dataset exhibits significant noise. Although the prediction accuracy for the activation energy Δh∗ and the pre-exponential factor log(A) was somewhat lower, the method still provides valuable estimates that can be refined with supplementary techniques.This work highlights the potential of machine learning approaches like MCNN to revolutionize the parameter extraction process in complex physical models, reducing the reliance on manual curve-fitting and providing a more automated, scalable solution. We also analyze the primary sources of prediction errors in the MCNN outputs and offer insights into future improvements, including model architecture refinements and the integration of additional physical constraints. Our findings suggest that this approach can be extended to other domains where similar models are employed, paving the way for broader applications of machine learning in materials science.

FedART: A neural model integrating federated learning and adaptive resonance theory

Federated Learning (FL) has emerged as a promising paradigm for collaborative model training across distributed clients while preserving data privacy. However, prevailing FL approaches aggregate the clients’ local models into a global model through multi-round iterative parameter averaging. This leads to the undesirable bias of the aggregated model towards certain clients in the presence of heterogeneous data distributions among the clients. Moreover, such approaches are restricted to supervised classification tasks and do not support unsupervised clustering. To address these limitations, we propose a novel one-shot FL approach called Federated Adaptive Resonance Theory (FedART) which leverages self-organizing Adaptive Resonance Theory (ART) models to learn category codes, where each code represents a cluster of similar data samples. In FedART, the clients learn to associate their private data with various local category codes. Under heterogeneity, the local codes across different clients represent heterogeneous data. In turn, a global model takes these local codes as inputs and aggregates them into global category codes, wherein heterogeneous client data is indirectly represented by distinctly encoded global codes, in contrast to the averaging out of parameters in the existing approaches. This enables the learned global model to handle heterogeneous data. In addition, FedART employs a universal learning mechanism to support both federated classification and clustering tasks. Our experiments conducted on various federated classification and clustering tasks show that FedART consistently outperforms state-of-the-art FL methods on data with heterogeneous distribution across clients.

Localization of the epileptogenic network from scalp EEG using a patient-specific whole-brain model

ABSTRACT Computational modeling is a key tool for elucidating the neuronal mechanisms underlying epileptic activity. Despite considerable progress, existing models often lack realistic accuracy in representing electrophysiological epileptic activity. In this study, we used a comprehensive human brain model based on a neural mass model, which is tailored to the layered structure of the neocortex and incorporates patient-specific imaging data. This approach allowed the simulation of scalp EEGs in an epileptic patient suffering from type 2 focal cortical dysplasia (FCD). The simulation specifically addressed epileptic activity induced by FCD, faithfully reproducing intracranial interictal epileptiform discharges (IEDs) recorded with electrocorticography. For constructing the patient-specific scalp EEG, we carefully defined a clear delineation of the epileptogenic zone by numerical simulations to ensure fidelity to the topography, polarity, and diffusion characteristics of IEDs. This nuanced approach improves the accuracy of the simulated EEG signal, provides a more accurate representation of epileptic activity, and enhances our understanding of the mechanism behind the epileptogenic networks. The accuracy of the model was confirmed by a postoperative reevaluation with a secondary EEG simulation that was consistent with the lesion’s removal. Ultimately, this personalized approach may prove instrumental in optimizing and tailoring epilepsy treatment strategies.

Magnetic sludge-derived biochar for Cu (II) removal: RSM-based preparation and BP artificial neural network adsorption modeling

Magnetic sludge-derived biochar (MSDB) was synthesized using response surface methodology (RSM) optimization, with sludge as the raw material and sodium pyrophosphate as the modifier. The structure and surface properties of MSDB were characterized using scanning electron microscopy (SEM), Fourier transform infrared spectroscopy (FTIR), X-ray diffraction (XRD), X-ray photoelectron spectroscopy (XPS), vibrating sample magnetometer (VSM), and Brunauer-Emmett-Teller (BET) analysis. These analyses revealed that MSDB is rich in functional groups and exhibits magnetic properties, facilitating the adsorption of Cu(II) via complexation and easy separation of MSDB through magnetic means. Batch adsorption experiments showed that under conditions of 25 °C and pH 3.0, with an MSDB dosage of 0.75 g L−1, a Cu(II) removal efficiency of 95.67 % was achieved. Thermodynamic studies demonstrated maximum adsorption capacities of 173.69 mg g−1 at 25 °C, 183.26 mg g−1 at 35 °C, and 197.96 mg g−1 at 45 °C. Additionally, the BP neural network model showed a high coefficient of determination (R2 = 0.9396) and a low mean squared error (MSE = 0.0268) in simulating the MSDB-Cu(II) adsorption process, confirming the robustness of the model in both data simulation and predictive accuracy. This method offers a promising approach for kinetic modeling and predicting adsorption behavior in related fields of study.

Inverse feedforward control of piezoelectric actuators using optimized composite neural network-based Hammerstein model

This paper utilizes the optimized composite neural network (OCNN)-Hammerstein model to directly identify the dynamic inverse hysteresis effect, employing it as a feedforward controller to compensate for the rate-dependent and amplitude-dependent dynamic hysteresis of the piezoelectric actuator. The OCNN-Hammerstein model consists of an optimized composite neural network and an auto-regressive exogenous model in series. The OCNN comprises convolutional neural network layer and radial basis function neural network layer, with its hyperparameters optimized using a modified grey wolf optimizer algorithm. Compared to the existing dynamic Prandtl-Ishlinskii-Hammerstein and dynamic Bouc-Wen-Hammerstein models in the literature, the feedforward controller based on the proposed model in this paper demonstrates superior performance, with an RSME below 0.45 μm and a 74.5% reduction in relative hysteresis at the condition of 300 Hz and maximum amplitude. The feedforward controller exhibits universality across all amplitude ranges and within the 50–300 Hz frequency range, while also providing predictive compensation for dynamic hysteresis at 300–400 Hz. The feedforward controller based on the OCNN-Hammerstein model in this paper provides a crucial methodology for open-loop control of piezoelectric actuators.

Automatic calculation of inclination angle of a structure based on 3D reconstruct digital twin using neural deep learning model and SFM technology

Automatic calculation of inclination angle of a structure based on 3D reconstruct digital twin using neural deep learning model and SFM technology

Random-Coupled Neural Network

Improving the efficiency of current neural networks and modeling them on biological neural systems have become prominent research directions in recent years. The pulse-coupled neural network (PCNN) is widely used to mimic the computational characteristics of the human brain in computer vision and neural network fields. However, PCNN faces limitations such as limited neural connections, high computational costs, and a lack of stochastic properties. This study proposes a random-coupled neural network (RCNN) to address these limitations. RCNN employs a stochastic inactivation process, selectively inactivating neural connections using a random inactivation weight matrix. This method reduces the computational burden and allows for extensive neural connections. RCNN encodes constant stimuli as periodic spike trains and periodic stimuli as chaotic spike trains, reflecting the information encoding characteristics of biological neural systems. Our experiments applied RCNN to image segmentation and fusion tasks, demonstrating its robustness, efficiency, and high noise resistance. Results indicate that RCNN surpasses traditional methods in performance across these applications.

An Optimal Fuzzy Neural Network Prediction Model for Student Performance Prediction in Online Education

An Optimal Fuzzy Neural Network Prediction Model for Student Performance Prediction in Online Education

Automatic detection of temporomandibular joint osteoarthritis radiographic features using deep learning artificial intelligence. A Diagnostic accuracy study

ObjectiveThe purpose of this study was to investigate the diagnostic performance of a neural network Artificial Intelligence model for the radiographic confirmation of Temporomandibular Joint Osteoarthritis in reference to an experienced radiologist. Materials and MethodsThe diagnostic performance of an AI model in identifying radiographic features in patients with TMJ-OA was evaluated in a diagnostic accuracy cohort study. Adult patients elected for radiographic examination by the Diagnostic Criteria for Temporomandibular Disorders decision tree were included. Cone-beam computed Tomography images were evaluated by object detection YOLO deep learning model. The diagnostic performance was verified against examiner radiographic evaluation. ResultsThe differences between the AI model and examiner were non-significant statistically, except in the subcortical cyst (P=0.049*). AI model showed substantial to near-perfect levels of agreement when compared to those of the examiner data. Regarding each radiographic phenotype, the AI model reported favorable sensitivity, specificity, accuracy, and highly statistically significant Receiver Operating Characteristic (ROC) analysis (p < 0.001). Area Under Curve ranged from 0.872, for surface erosion, to 0.911 for subcortical cyst. ConclusionAI object detection model could open the horizon for a valid, automated, and convenient modality for TMJ-OA radiographic confirmation and radiomic features identification with a significant diagnostic power.

Predicting an opaque bubble layer during small-incision lenticule extraction surgery based on deep learning.

This study aimed to predict the formation of OBL during femtosecond laser SMILE surgery by employing deep learning technology. This was a cross-sectional, retrospective study conducted at a university hospital. Surgical videos were randomly divided into a training (3,271 patches, 73.64%), validation (704 patches, 15.85%), and internal verification set (467 patches, 10.51%). An artificial intelligence (AI) model was developed using a SENet-based residual regression deep neural network. Model performance was assessed using the mean absolute error (E MA ), Pearson's correlation coefficient (r), and determination coefficient (R 2 ). Four distinct types of deep neural network models were established. The modified deep residual neural network prediction model with channel attention built on the PyTorch framework demonstrated the best predictive performance. The predicted OBL area values correlated well with the Photoshop-based measurements (E MA = 0.253, r = 0.831, R 2 = 0.676). The ResNet (E MA = 0.259, r = 0.798, R 2 = 0.631) and Vgg19 models (E MA = 0.31, r = 0.758, R 2 = 0.559) both displayed satisfactory predictive performance, while the U-net model (E MA = 0.605, r = 0.331, R 2 = 0.171) performed poorest. We used a panoramic corneal image obtained before the SMILE laser scan to create a unique deep residual neural network prediction model to predict OBL formation during SMILE surgery. This model demonstrated exceptional predictive power, suggesting its clinical applicability across a broad field.

The Enhancement of Machine Learning-Based Engine Models Through the Integration of Analytical Functions

The integration of analytical functions into machine learning-based engine models represents a significant advancement in predictive performance and operational efficiency. This paper focuses on the development of hybrid approaches to model engine combustion and temperature indices and on the synergistic effects of combining traditional analytical methods with modern machine learning techniques (such as artificial neural networks) to enhance the accuracy and robustness of such models. The main innovative contribution of this paper is the integration of analytical functions to improve the extrapolation capabilities of the data-driven models. In this work, it is demonstrated that the integrated models achieve superior predictive accuracy and generalization performance across dynamic engine operating conditions, with respect to purely neural network-based models. Furthermore, the analytical corrective functions force the output of the complete model to follow a physical trend and to assume consistent values also outside the domain of values assumed by the input features during the training procedure of the neural networks. This study highlights the potential of this integrative approach based on the implementation of the effects superposition principle. Such an approach also allows us to solve one of the intrinsic issues of data-driven modeling, without increasing the complexity of the training data’s collection and pre-processing.

A Plantar Pressure Detection and Gait Analysis System Based on Flexible Triboelectric Pressure Sensor Array and Deep Learning.

Gait detection is essential for the assessment of human health status and early diagnosis of diseases. The current gait analysis systems are bulky, limited in the scope of use, and cause interference with the movement of the measured person. Hence, it is necessary to develop a wearable gait detection system that is soft, breathable, lightweight, and self-powered. Here, a plantar pressure sensor array and gait analysis system based on a flexible triboelectric pressure sensor (FTPS) array is developed. Soft, breathable, and wearable electrospinning nanofiber film with excellent triboelectric properties is used as the plantar pressure sensor, achieving a high sensitivity of 45.1 mVkPa-1 in the range of 40-200 kPa and 19.4 mVkPa-1 in the range of 200-400 kpa. 32 FTPSs are integrated into an intelligent insole, which has the characteristics of soft, easy production, good air permeability, long-time stability, no external power supply, and etc. Based on the long short-term memory artificial neural network deep learning model, the accuracy of gait judgment can reach 94.23%. This work provides a feasible solution for real-time gait detection, which will have potential applications in human health assessment and early diagnosis of disease.

Enhancing 3D planetary atmosphere simulations with a surrogate radiative transfer model

Abstract This work introduces an approach to enhancing the computational efficiency of 3D atmospheric simulations by integrating a machine-learned surrogate model into the OASIS global circulation model (GCM). Traditional GCMs, which are based on repeatedly numerically integrating physical equations governing atmospheric processes across a series of time-steps, are time-intensive, leading to compromises in spatial and temporal resolution of simulations. This research improves upon this limitation, enabling higher resolution simulations within practical timeframes. Speeding up 3D simulations holds significant implications in multiple domains. Firstly, it facilitates the integration of 3D models into exoplanet inference pipelines, allowing for robust characterisation of exoplanets from a previously unseen wealth of data anticipated from JWST and post-JWST instruments. Secondly, acceleration of 3D models will enable higher resolution atmospheric simulations of Earth and Solar System planets, enabling more detailed insights into their atmospheric physics and chemistry. Our method replaces the radiative transfer module in OASIS with a recurrent neural network-based model trained on simulation inputs and outputs. Radiative transfer is typically one of the slowest components of a GCM, thus providing the largest scope for overall model speed-up. The surrogate model was trained and tested on the specific test case of the Venusian atmosphere, to benchmark the utility of this approach in the case of non-terrestrial atmospheres. This approach yields promising results, with the surrogate-integrated GCM demonstrating above 99.0% accuracy and 147 factor GPU speed-up of the entire simulation compared to using the matched original GCM under Venus-like conditions.

Machine learning-driven discovery of high-performance MEMS disk resonator gyroscope structural topologies

The design of the microelectromechanical system (MEMS) disc resonator gyroscope (DRG) structural topology is crucial for its physical properties and performance. However, creating novel high-performance MEMS DRGs has long been viewed as a formidable challenge owing to their enormous design space, the complexity of microscale physical effects, and time-consuming finite element analysis (FEA). Here, we introduce a new machine learning-driven approach to discover high-performance DRG topologies. We represent the DRG topology as pixelated binary matrices and formulate the design task as a path-planning problem. This path-planning problem is solved via deep reinforcement learning (DRL). In addition, we develop a convolutional neural network-based surrogate model to replace the expensive FEA to provide reward signals for DRL training. Benefiting from the computational efficiency of neural networks, our approach achieves a significant acceleration ratio of 4.03 × 105 compared with FEA, reducing each DRL training run to only 426.5 s. Through 8000 training runs, we discovered 7120 novel structural topologies that achieve navigation-grade precision. Many of these surpass traditional designs in performance by several orders of magnitude, revealing innovative solutions previously unconceived by humans.

Artificial Neural Network-based Model for Predicting Cardiologists’ Over-apron Dose in CATHLABs

Abstract Aim: The radiation dose that cardiologists receive in the catheterization laboratory is influenced by various factors. Handling high-stress tasks in interventional cardiology departments may cause physicians to overlook the use of dosimeters. Therefore, it is essential to develop a model for predicting cardiologists’ radiation exposure. Materials and Methods: This study developed an artificial neural network (ANN) model to predict the over-apron radiation dose received by cardiologists during catheterization procedures, using dose area product (DAP) values. Leveraging a validated Monte Carlo simulation program, we generated data from simulations with varying spectra (70, 81, and 90 kVp) and tube orientations, resulting in 125 unique scenarios. We then used these data to train a multilayer perceptron neural network with four input features: DAP, energy spectrum, tube angulation, and the resulting cardiologist’s dose. Results: The model demonstrated high predictive accuracy with a correlation coefficient (R-value) of 0.95 and a root mean square error (RMSE) of 3.68 μSv, outperforming a traditional linear regression model, which had an R-value of 0.48 and an RMSE of 18.15 μSv. This significant improvement highlights the effectiveness of advanced techniques such as ANNs in accurately predicting occupational radiation doses. Conclusion: This study underscores the potential of ANN models for accurate radiation dose prediction, enhancing safety protocols, and providing a reliable tool for real-time exposure assessment in clinical settings. Future research should focus on broader validation and integration into real-time monitoring systems.

Assessing the Profit Impact of ARIMA and Neural Network Demand Forecasts in Retail Inventory Replenishment

This study investigates the integration of demand forecasting and inventory replenishment strategies to enhance retail profitability. A deterministic optimal replenishment model is utilized to analyze the predictive performance of various neural network architectures and ARIMA models using real sales data. The predictive accuracy and subsequent influence on optimal firm profits over a multi-period planning horizon is assessed. The Integer Programming model devised optimizes daily replenishment across multiple retail routes, taking into account sales revenue, supply costs, inventory holding, sales loss, and transportation expenses. The study is distinctive in its dual assessment: it evaluates both the accuracy of forecasting methods and their direct impact on profitability through systematic inventory decisions. Neural network architectures selected for minimizing error in product sales predictions have 6% lower mean squared error compared to Akaike Information Criterion minimizing ARIMA models. For longer horizon predictions necessary in performance gap grows larger, e.g., with %60 difference for predictions 15 days ahead. Predictions reflect as 1.6% higher profits on average, when neural network predictions and more efficient longer planning horizons of the optimization model are preferred. Planning 30 days ahead, optimizing with neural network predictions elicits 2.3% higher profits compared to profits attainable based on ARIMA predictions. Our findings illustrate how different forecasting methods can affect firm profitability by shaping inventory replenishment strategies. By merging mathematical optimization with time series forecasting, this research provides a comprehensive evaluation of how advanced predictive technologies can enhance retail inventory practices and improve profitability.

Seizure Sources Can Be Imaged from Scalp EEG by Means of Biophysically Constrained Deep Neural Networks.

Seizure localization is important for managing drug-resistant focal epilepsy. Here, the capability of a novel deep learning-based source imaging framework (DeepSIF) for imaging seizure activities from electroencephalogram (EEG) recordings in drug-resistant focal epilepsy patients is investigated. The neural mass model of ictal oscillations is adopted to generate synthetic training data with spatio-temporal-spectra features similar to ictal dynamics. The trained DeepSIF model is rigorously validated using computer simulations and in a cohort of 33 drug-resistant focal epilepsy patients with high-density (76-channel) EEG seizure recordings, by comparing DeepSIF estimates with surgical resection outcome and seizure onset zone (SOZ) . These findings show that the trained DeepSIF model outperforms other methods in estimating the spatial and temporal information of origins of ictal activities. It achieves a high spatial specificity of 96% and a low spatial dispersion of 3.80±5.74mm when compared to the resection region. The source imaging results also demonstrate good coverage of SOZ, with an average distance of 10.89±10.14mm (from the SOZ to the reconstruction). These promising results suggest that DeepSIF has significant potential for advancing noninvasive imaging of the origins of ictal activities in patients with focal epilepsy, aiding management of intractable epilepsy.

Artificial Neural Network Surrogate Model for Geochemical Calculations in Pore-Scale Reactive Transport Simulations

Artificial Neural Network Surrogate Model for Geochemical Calculations in Pore-Scale Reactive Transport Simulations