Neural Network Research Articles

Clinical prediction model for MODY type diabetes mellitus in children

BACKGROUND: MODY (maturity-onset diabetes of the young) is a rare monogenic form of diabetes mellitus, the gold standard of diagnosis is mutations detection in the genes responsible for the development of this form diabetes. Genetic test is expensive and takes a lot of time. The diagnostic criteria for MODY are well known. The development of clinical decision support system (CDSS) which allows physicians based on clinical data to determine who should have molecular genetic testing is relevant.AIM: Provided a retrospective analysis of clinical data of the patients with T1DM and MODY, from 0 to 18 years old, regardless of the duration of the disease to develop the model. Based on clinical data, a feedforward neural network (NN) was implemented - a multilayer perceptron.MATERIALS AND METHODS: Development of the most effective algorithm for predicting MODY in children based on available clinical indicators of 1710 patients with diabetes under the age of 18 years using a multilayer feedforward neural network.RESULTS: The sample consisted of 1710 children under the age of 18 years with T1DM (78%) and MODY (22%) diabetes. For the final configuration of NS the following predictors were selected: gender, age at passport age, age at the diagnosis with DM, HbA1c, BMI SDS, family history of DM, treatment. The performance (quality) assessment of the NN was carried out on a test sample (the area under the ROC (receiver operating characteristics) curve reached 0.97). The positive predictive value of PCPR was achieved at a cut-off value of 0.40 (predicted probability of MODY diabetes 40%). At which the sensitivity was 98%, specificity 93%, PCR with prevalence correction was 78%, and PCR with prevalence correction was 99%, the overall accuracy of the model was 94%.Based on the NN model, a CDSS was developed to determine whether a patient has MODY diabetes, implemented as an application.CONCLUSION: The clinical prediction model MODY developed in this work based on the NN, uses the clinical characteristic available for each patient to determine the probability of the patient having MODY. The use of the developed model in clinical practice will assist in the selection of patients for diagnostic genetic testing for MODY, which will allow for the efficient allocation of healthcare resources, the selection of personalized treatment and patient monitoring.

Accurate wire rope defect MFL detection using improved Hilbert transform and LSTM neural network

ABSTRACT Wire rope is playing a significant role in various engineering fields. Although a great number of non-destructive testing methods are applied to wire rope defect detection, the results are greatly influenced by the signal noises and the quantitative defect detection of wire rope is consequently limited. Therefore, a new accurate and quantitative wire rope defect magnetic flux leakage (MFL) recognition method based on improved Hilbert transform and long short-term memory (LSTM) neural network is proposed. The theoretical background of the proposed signal processing models and preliminary results, as well as the influence of the main model parameters are analyzed first. Then, different wire rope defect classification principles through LSTM neural network and performance evaluation based on the improved signal processing and machine learning combined method are presented. Finally, the defect inspection and classification comparison results between the proposed method and other deep learning neural networks and pattern recognition strategies are given, which manifest that the proposed method has higher classification accuracy and shorter running time for five different wire rope defects under various servicing conditions, and is promising in accurate defect inspection for multi-scenario wire rope. Additionally, the main conclusions, disadvantages of the proposed method are summarized.

Genotype sampling for deep-learning assisted experimental mapping of a combinatorially complete fitness landscape.

Experimental characterization of fitness landscapes, which map genotypes onto fitness, is important for both evolutionary biology and protein engineering. It faces a fundamental obstacle in the astronomical number of genotypes whose fitness needs to be measured for any one protein. Deep learning may help to predict the fitness of many genotypes from a smaller neural network training sample of genotypes with experimentally measured fitness. Here I use a recently published experimentally mapped fitness landscape of more than 260,000 protein genotypes to ask how such sampling is best performed. I show that multilayer perceptrons, recurrent neural networks (RNNs), convolutional networks, and transformers, can explain more than 90 percent of fitness variance in the data. In addition, 90 percent of this performance is reached with a training sample comprising merely ≈103 sequences. Generalization to unseen test data is best when training data is sampled randomly and uniformly, or sampled to minimize the number of synonymous sequences. In contrast, sampling to maximize sequence diversity or codon usage bias reduces performance substantially. These observations hold for more than one network architecture. Simple sampling strategies may perform best when training deep learning neural networks to map fitness landscapes from experimental data. The fitness landscape data analyzed here is publicly available as described previously (Papkou, et al., 2023). All code used to analyze this landscape is publicly available at https://github.com/andreas-wagner-uzh/fitness_landscape_sampling. Supplementary data are available at Bioinformatics online.

Ancient mural dynasty recognition algorithm based on a neural network architecture search

A neural network model needs to be manually designed for ancient mural dynasty recognition, and this paper proposes an ancient mural dynasty recognition algorithm that is based on a neural architecture search (NAS). First, the structural edge information of mural images is extracted for use by the neural network model in recognizing mural missions. Second, an NAS algorithm that is based on contrast selection (CS) simplifies the architecture search to an incremental CS and then searches for the optimal network architecture on the mural dataset. Finally, the identified optimal network architecture is used for training and testing to complete the mural dynasty recognition task. The results show that the top accuracy of the proposed method on the mural dataset is 88.10%, the recall rate is 87.52%, and the precision rate is 87.69%. Each evaluation index used by the neural network model is superior to that of classical network models such as AlexNet and ResNet-50. Compared with NAS methods such as ASNG and MIGO, the accuracy of mural dynasty recognition is higher by an average of 4.27% when using the proposed method. The proposed method is verified on CIFAR-10, CIFAR-100, ImageNet16-120 and other datasets and achieves a good recognition accuracy in the NAS-bench-201 search space, which averages 93.26%, 70.73% and 45.34%, respectively, on the abovementioned datasets.

Retrosynthesis Zero: Self-Improving Global Synthesis Planning Using Reinforcement Learning.

The field of computer-aided synthesis planning (CASP) has witnessed significant growth in recent years. Still, many CASP programs rely on large data sets to train neural networks, resulting in limitations due to the data quality and prior knowledge from chemists. In response, we propose Retrosynthesis Zero (ReSynZ), a reaction template-based method that combines Monte Carlo Tree Search with reinforcement learning inspired by AlphaGo Zero. Unlike other single-step reaction template-based CASP methods, ReSynZ takes complete synthesis paths for complex molecules, determined by reaction rules, as input for training the neural network. ReSynZ enables neural networks trained with relatively small reaction data sets (tens of thousands of data) to generate multiple synthesis pathways for a target molecule and suggest possible reaction conditions. On multiple data sets of molecular retrosynthesis, ReSynZ demonstrates excellent predictive performance compared to existing algorithms. The advantages, such as self-improving model features, flexible reward settings, the potential to surpass human limitations in chemical synthesis route planning, and others, make ReSynZ a valuable tool in chemical synthesis design.

Residual network improves the prediction accuracy of genomic selection.

Genetic improvement of complex traits in animal and plant breeding depends on the efficient and accurate estimation of breeding values. Deep learning methods have been shown to be not superior over traditional genomic selection (GS) methods, partially due to the degradation problem (i.e. with the increase of the model depth, the performance of the deeper model deteriorates). Since the deep learning method residual network (ResNet) is designed to solve gradient degradation, we examined its performance and factors related to its prediction accuracy in GS. Here we compared the prediction accuracy of conventional genomic best linear unbiased prediction, Bayesian methods (BayesA, BayesB, BayesC, and Bayesian Lasso), and two deep learning methods, convolutional neural network and ResNet, on three datasets (wheat, simulated and real pig data). ResNet outperformed other methods in both Pearson's correlation coefficient (PCC) and mean squared error (MSE) on the wheat and simulated data. For the pig backfat depth trait, ResNet still had the lowest MSE, whereas Bayesian Lasso had the highest PCC. We further clustered the pig data into four groups and, on one separated group, ResNet had the highest prediction accuracy (both PCC and MSE). Transfer learning was adopted and capable of enhancing the performance of both convolutional neural network and ResNet. Taken together, our findings indicate that ResNet could improve GS prediction accuracy, affected potentially by factors such as the genetic architecture of complex traits, data volume, and heterogeneity.

Black-Box Attacks on Graph Neural Networks via White-Box Methods With Performance Guarantees

Black-Box Attacks on Graph Neural Networks via White-Box Methods With Performance Guarantees

Borich 요구도와 The Locus for Focus model을 활용한 대학생의 챗GPT 현 수준 및 중요도 분석

Objectives The purpose of this study is to analyze the perception level of university students regarding ChatGPT using the Borich Needs and The Locus for Focus model. Methods For this purpose, a total of 199 undergraduate students from University A in the metropolitan area were selected as participants, and the research was conducted using SPSS 25.0 and Excel software programs. Statistical significance was analyzed using t-tests to examine the differences between the current level and im-portance in IPA analysis.Borich'sNeeds and The Locus for Focus Model were examined in a 2x2 matrix, divided into four quadrants. Results The key findings of this study are as follows: First, in the IPA analysis, the reinforcement and maintenance area(HH) revealed 2 items related to understanding ChatGPT, 5items related to ChatGPT principles and applica-tions, and 1 item related to data and machine learning. The concentration management area(HL) included 1 item related to ChatGPT principles and applications, and 3 items related to data and machine learning. The low priority area(LL) consisted of 1 item related to understanding ChatGPT, 2 items related to ChatGPT principles and applica-tions, and 4 items related to social impact. The excessive area(LH) was characterized bybiaseddata in the social impact category. Second, it was found that the differences between importance and current level were higher in terms of importance for all items, except for those related to data bias. This difference was statistically significant. The priority order according to Borich's needs includes 'structuredand unstructured data', 'sensorsand perception', 'computer vision', 'concepts of AI agents' and 'conceptsand application learning of machines'. Third, The Locus for Focus model analysis revealed that the HH area consisted of 9 items, including 'speech recognition and language understanding', 'problem solving and exploration', 'representation and deduction of information', 'concepts and ap-plication learning of machines', 'concepts and application of artificial neural networks', 'data attributes', 'structured and unstructured data', 'classification models', and 'machine learning model implementation'. Fourth, a total of 6 items, including 'speech recognition and language understanding', 'representation and deduction of information', 'concepts and application learning of machines', 'concepts and application of artificial neural networks', 'structured and unstructured data', and 'classification models' were consistently rated as high priority and demanded the high-est attention. Conclusions The findings of this study indicate the need to explore ways to apply ChatGPT in education. Furthermore, it proposes a long-term education plan to enhance awareness of the social impact of ChatGPT and provide practical strategies for learning and applying it.

Programmable Tanh- and ELU-Based Photonic Neurons in Optics-Informed Neural Networks

Programmable Tanh- and ELU-Based Photonic Neurons in Optics-Informed Neural Networks

Research on Double Encryption of Ghost Imaging by SegNet Deep Neural Network

Research on Double Encryption of Ghost Imaging by SegNet Deep Neural Network

Physiologic validation of the compensatory reserve metric obtained from pulse oximetry: A step towards advanced medical monitoring on the battlefield.

The Compensatory Reserve Metric (CRM) provides a time sensitive indicator of hemodynamic decompensation. However, its in-field utility is limited due to the size and cost-intensive nature of standard vital sign monitors or photoplethysmographic volume-clamp (PPGVC) devices used to measure arterial waveforms. In this regard, photoplethysmographic measurements obtained from pulse oximetry (PPGPO) may serve as a useful, portable alternative. This study aimed to validate CRM values obtained using PPGPO. Forty-nine healthy adults (25 females) underwent a graded lower body negative pressure (LBNP) protocol to simulate hemorrhage. Arterial waveforms were sampled using PPGPO and PPGVC. The CRM was calculated using a one-dimensional convolutional neural network. Cardiac output and stroke volume were measured using PPGVC. A brachial artery catheter was used to measure intraarterial pressure. A 3-lead ECG was used to measure heart rate. Fixed-effect linear mixed models with repeated measures were used to examine the association between CRM values and physiologic variables. Log-rank analyses were used to examine differences in shock determination during LBNP between monitored hemodynamic parameters. The median LBNP stage reached was 70 mmHg (Range: 45-100 mmHg). Relative to baseline, at tolerance there was a 47±12% reduction in stroke volume, 64±27% increase in heart rate, and 21±7% reduction in systolic blood pressure (P<0.001 for all). CRM values obtained with both PPGPO and PPGVC were associated with changes in heart rate (P<0.001), stroke volume (P<0.001), and pulse pressure (P<0.001). Furthermore, they provided an earlier detection of hemodynamic shock relative to the traditional metrics of shock index (P<0.001 for both), systolic blood pressure (P<0.001 for both), and heart rate (P=0.001 for both). The CRM obtained from PPGPO provides a valid, time-sensitized prediction of hemodynamic decompensation, opening the door to provide military medical personnel noninvasive in-field advanced capability for early detection of hemorrhage and imminent onset of shock. Diagnostic Tests or Criteria, Level IV.

Advancements in machine learning for predicting phases in high-entropy alloys: a comprehensive review

High entropy alloys (HEAs) are distinguished by their enhanced physicochemical properties, attributed to the formation of various phases such as solid solution (SS), intermetallic (IM), or a combination (SS + IM). These phases contribute distinctively to the microstructure of the alloys. A critical aspect of alloy design revolves around accurately predicting these phases, which has led to the integration of sophisticated data vetting methods and Machine Learning (ML) algorithms in recent research. This review paper aims to provide a comprehensive analysis of the advancements in phase prediction accuracy within HEAs, an essential component in the development of these alloys. HEAs are known for their intricate compositions, offering a wide spectrum of material properties, making them particularly relevant for applications aimed at future sustainability. Phase engineering in HEAs unlocks the potential for creating materials tailored to eco-friendly technologies and energy-efficient solutions. The challenge in predicting phase selection in HEAs is accentuated by the limited data available on these complex materials. This review delves into how advanced data vetting techniques and ML algorithms are being employed to overcome these challenges, thus contributing significantly to sustainable material design. The paper examines various algorithms used in HEA phase prediction, including KNN (K-Nearest Neighbors), SVM (Support Vector Machines), ANN (Artificial Neural Networks), GNB (Gaussian Naive Bayes), and RF (Random Forest). It discusses the testing accuracy of these algorithms in classifying HEA phases, revealing variations in their effectiveness. The review highlights the superior accuracy of ANNs, followed closely by KNN and SVM, while noting the comparatively lower accuracy of GNB. This comprehensive review synthesizes current research efforts in utilizing computational methods to design HEAs, underlining their broader implications in expediting the discovery and development of diverse metal alloys. These efforts are pivotal in meeting the evolving demands of modern engineering applications, thereby contributing to the advancement of materials science.

Integration of computer networks and artificial neural networks for an AI-based network operator

Integration of computer networks and artificial neural networks for an AI-based network operator

Hierarchical Training of Deep Neural Networks Using Early Exiting.

Deep neural networks (DNNs) provide state-of-the-art accuracy for vision tasks, but they require significant resources for training. Thus, they are trained on cloud servers far from the edge devices that acquire the data. This issue increases communication cost, runtime, and privacy concerns. In this study, a novel hierarchical training method for DNNs is proposed that uses early exits in a divided architecture between edge and cloud workers to reduce the communication cost, training runtime, and privacy concerns. The method proposes a brand-new use case for early exits to separate the backward pass of neural networks between the edge and the cloud during the training phase. We address the issues of most available methods that, due to the sequential nature of the training phase, cannot train the levels of hierarchy simultaneously or they do it with the cost of compromising privacy. In contrast, our method can use both edge and cloud workers simultaneously, does not share the raw input data with the cloud, and does not require communication during the backward pass. Several simulations and on-device experiments for different neural network architectures demonstrate the effectiveness of this method. It is shown that the proposed method reduces the training runtime for VGG-16 and ResNet-18 architectures by 29% and 61% in CIFAR-10 classification and by 25% and 81% in Tiny ImageNet classification, respectively, when the communication with the cloud is done over a low bit rate channel. This gain in the runtime is achieved, while the accuracy drop is negligible. This method is advantageous for online learning of high-accuracy DNNs on sensor-holding low-resource devices such as mobile phones or robots as a part of an edge-cloud system, making them more flexible in facing new tasks and classes of data.

Effects of local reduction of endogenous α-synuclein using antisense oligonucleotides on the fibril-induced propagation of pathology through the neural network in wild-type mice

In Parkinson's disease and other synucleinopathies, fibrillar forms of α-synuclein (aSyn) are hypothesized to structurally convert and pathologize endogenous aSyn, which then propagates through the neural connections, forming Lewy pathologies and ultimately causing neurodegeneration. Inoculation of mouse-derived aSyn preformed fibrils (PFFs) into the unilateral striatum of wild-type mice causes widespread aSyn pathologies in the brain through the neural network. Here, we used the local injection of antisense oligonucleotides (ASOs) against Snca mRNA to confine the area of endogenous aSyn protein reduction and not to affect the PFFs properties in this model. We then varied the timing and location of ASOs injection to examine their impact on the initiation and propagation of aSyn pathologies in the whole brain and the therapeutic effect using abnormally-phosphorylated aSyn (pSyn) as an indicator. By injecting ASOs before or 0–14 days after the PFFs were inoculated into the same site in the left striatum, the reduction in endogenous aSyn in the striatum leads to the prevention and inhibition of the regional spread of pSyn pathologies to the whole brain including the contralateral right hemisphere. ASO post-injection inhibited extension from neuritic pathologies to somatic ones. Moreover, injection of ASOs into the right striatum prevented the remote regional spread of pSyn pathologies from the left striatum where PFFs were inoculated and no ASO treatment was conducted. This indicated that the reduction in endogenous aSyn protein levels at the propagation destination site can attenuate pSyn pathologies, even if those at the propagation initiation site are not inhibited, which is consistent with the original concept of prion-like propagation that endogenous aSyn is indispensable for this regional spread. Our results demonstrate the importance of recruiting endogenous aSyn in this neural network propagation model and indicate a possible potential for ASO treatment in synucleinopathies.

Accelerated Gradient Approach For Deep Neural Network-Based Adaptive Control of Unknown Nonlinear Systems.

Recent connections in the adaptive control literature to continuous-time analogs of Nesterov's accelerated gradient method have led to the development of new real-time adaptation laws based on accelerated gradient methods. However, previous results assume that the system's uncertainties are linear-in-the-parameters (LIP). To compensate for non-LIP uncertainties, our preliminary results developed a neural network (NN)-based accelerated gradient adaptive controller to achieve trajectory tracking for nonlinear systems; however, the development and analysis only considered single-hidden-layer NNs. In this article, a generalized deep NN (DNN) architecture with an arbitrary number of hidden layers is considered, and a new DNN-based accelerated gradient adaptation scheme is developed to generate estimates of all the DNN weights in real-time. A nonsmooth Lyapunov-based analysis is used to guarantee the developed accelerated gradient-based DNN adaptation design achieves global asymptotic tracking error convergence for general nonlinear control affine systems subject to unknown (non-LIP) drift dynamics and exogenous disturbances. A comprehensive set of simulation studies are conducted on a two-state nonlinear system, a robotic manipulator, and a complex 20-D nonlinear system to demonstrate the improved performance of the developed method. Our simulation studies demonstrate enhanced tracking and function approximation performance from both DNN architectures and accelerated gradient adaptation.

Retraction Note: Reality of virtual damage identification based on neural networks and vibration analysis of a damaged bridge under a moving vehicle

Retraction Note: Reality of virtual damage identification based on neural networks and vibration analysis of a damaged bridge under a moving vehicle

Retraction Note: Virtual reality of interior noises of vehicles based on boundary element and neural networks

Retraction Note: Virtual reality of interior noises of vehicles based on boundary element and neural networks

Modeling of Vehicle Mobility in Shallow Water With Data-Driven Hydrodynamics Model

Abstract In this study, a data-driven hydrodynamics model is proposed to enable quick prediction of vehicle mobility in shallow water, considering the effect of tire–soil interaction. To this end, a high-fidelity coupled vehicle–water interaction model using computational fluid dynamics (CFD) and multibody dynamics (MBD) solvers is developed to characterize the hydrodynamic loads exerted on a vehicle operated in shallow water, and it is used to generate training data for the data-driven hydrodynamics model. To account for the history-dependent hydrodynamic behavior, a long short-term memory (LSTM) neural network is introduced to incorporate effects of the historical variation of vehicle motion states as the input to the data-driven model, and it is used to predict hydrodynamic loads online exerted on vehicle components in the MBD mobility simulation. The impacts of hydrodynamic loads on the vehicle mobility capability in shallow water are examined for different water depths and incoming flow speeds using the high-fidelity coupled CFD-MBD model. Furthermore, it is demonstrated that the vehicle–water interaction behavior in scenarios not considered in the training data can be predicted using the proposed LSTM data-driven hydrodynamics model. However, the use of non-LSTM layers, which do not account for the sequential variation of vehicle motion states as the input, leads to an inaccurate prediction. A substantial computational speedup is achieved with the proposed LSTM-MBD vehicle–water interaction model while ensuring accuracy, compared to the computationally expensive high-fidelity coupled CFD-MBD model.

The use of artificial intelligence for predicting postinfarction myocardial viability in echocardiographic images.

Evaluation of standard echocardiographic examination with artificial intelligence may help in the diagnosis of myocardial viability and function recovery after acute coronary syndrome. Sixty-one consecutive patients with acute coronary syndrome were enrolled in the present study (43 men, mean age 61 ± 9 years). All patients underwent percutaneous coronary intervention (PCI). 533 segments of the heart echo images were used. After 12 ± 1 months of follow-up, patients had an echocardiographic evaluation. After PCI each patient underwent cardiac magnetic resonance (CMR) with late enhancement and low-dose dobutamine echocardiographic examination. For texture analysis, custom software was used (MaZda 5.20, Institute of Electronics).Linear and non-linear (neural network) discriminative analyses were performed to identify the optimal analytic method correlating with CMR regarding the necrosis extent and viability prediction after follow-up. Texture parameters were analyzed using machine learning techniques: Artificial Neural Networks, Namely Multilayer Perceptron, Nonlinear Discriminant Analysis, Support Vector Machine, and Adaboost algorithm. The mean concordance between the CMR definition of viability and three classification models in Artificial Neural Networks varied from 42% to 76%. Echo-based detection of non-viable tissue was more sensitive in the segments with the highest relative transmural scar thickness: 51-75% and 76-99%. The best results have been obtained for images with contrast for red and grey components (74% of proper classification). In dobutamine echocardiography, the results of appropriate prediction were 67% for monochromatic images. Detection and semi-quantification of scar transmurality are feasible in echocardiographic images analyzed with artificial intelligence. Selected analytic methods yielded similar accuracy, and contrast enhancement contributed to the prediction accuracy of myocardial viability after myocardial infarction in 12 months of follow-up.