Neural Network Model Research Articles

Why artificial intelligence needs sociology of knowledge: parts I and II

AbstractRecent developments in artificial intelligence based on neural nets—deep learning and large language models which together I refer to as NEWAI—have resulted in startling improvements in language handling and the potential to keep up with changing human knowledge by learning from the internet. Nevertheless, examples such as ChatGPT, which is a ‘large language model’, have proved to have no moral compass: they answer queries with fabrications with the same fluency as they provide facts. I try to explain why this is, basing the argument on the sociology of knowledge, particularly social studies of science, notably ‘studies of expertise and experience’ and the ‘fractal model’ of society. Learning from the internet is not the same as socialisation: NEWAI has no primary socialisation such as provides the foundations of human moral understanding. Instead, large language models are retrospectively socialised by human intervention in an attempt to align them with societally accepted ethics. Perhaps, as technology advances, large language models could come to understand speech and recognise objects sufficiently well to acquire the equivalent of primary socialisation. In the meantime, we must be vigilant about who is socialising them and be aware of the danger of their socialising us to align with them rather than vice-versa, an eventuality that would lead to the further erosion of the distinction between the true and the false giving further support to populism and fascism.

Risk analysis of noise-induced hearing loss of workers in the automobile manufacturing industries based on back-propagation neural network model: a cross-sectional study in Han Chinese population.

This study aims to predict the risk of noise-induced hearing loss (NIHL) through a back-propagation neural network (BPNN) model. It provides an early, simple and accurate prediction method for NIHL. Population based, a cross sectional study. Han, China. This study selected 3266 Han male workers from three automobile manufacturing industries. Information including personal life habits, occupational health test information and occupational exposure history were collected and predictive factors of NIHL were screened from these workers. BPNN and logistic regression models were constructed using these predictors. The input variables of BPNN model were 20, 16 and 21 important factors screened by univariate, stepwise and lasso-logistic regression. When the BPNN model was applied to the test set, it was found to have a sensitivity (TPR) of 83.33%, a specificity (TNR) of 85.92%, an accuracy (ACC) of 85.51%, a positive predictive value (PPV) of 52.85%, a negative predictive value of 96.46% and area under the receiver operating curve (AUC) is: 0.926 (95% CI: 0.891 to 0.961), which demonstrated the better overall properties than univariate-logistic regression modelling (AUC: 0.715) (95% CI: 0.652 to 0.777). The BPNN model has better predictive performance against NIHL than the stepwise-logistic and lasso-logistic regression model in terms of TPR, TNR, ACC, PPV and NPV (p<0.05); the area under the receiver operating characteristics curve of NIHL is also higher than that of the stepwise and lasso-logistic regression model (p<0.05). It was a relatively important factor in NIHL to find cumulative noise exposure, auditory system symptoms, age, listening to music or watching video with headphones, exposure to high temperature and noise exposure time in the trained BPNN model. The BPNN model was a valuable tool in dealing with the occupational risk prediction problem of NIHL. It can be used to predict the risk of an individual NIHL.

An integrated development environment based situational awareness for operational reliability evaluation in wind energy systems incorporating uncertainties

Wind power fluctuations impact power system flexibility and reliability, emphasizing situational awareness (SA) in wind power integration. Monitoring wind speed and power generation forms the core of SA, enabling insight into wind system conditions. An Integrated Development Environment (IDE) becomes pivotal in assessing the SA of wind-integrated systems, which is vital for forecasting operational reliability in power system planning. The IDE facilitates Simulink-based wind model development and is linked to dSPACE hardware through a real-time interface (RTI) for validation. This IDE-driven integration utilizes cloud-based Internet of Things (IoT) capabilities, acquiring wind speed and power generation data for precise SA evaluations. A multi-state wind speed model is employed to incorporate energy generation fluctuations. To improve this, the artificial neural network (ANN) model, i.e., a time series-based non-linear autoregressive with exogenous input (NARX) ANN, is utilized to predict future wind speed and power generation, strengthening SA-based operational reliability. The study investigates the effectiveness of a suggested operational planning strategy under wind generation uncertainties. This assessment is carried out within a real-time deployment of wind energy systems seamlessly integrated with IoT on the dSPACE platform. The primary objective is to evaluate the system's ability to maintain SA effectively.

Novel immune classification based on machine learning of pathological images predicts early recurrence of hepatocellular carcinoma

IntroductionImmune infiltration within the tumor microenvironment (TME) plays a significant role in the onset and progression of hepatocellular carcinoma (HCC). Machine learning applied to pathological images offers a practical means to explore the TME at the cellular level. Our former research employed a transfer learning procedure to adapt a convolutional neural network (CNN) model for cell recognition, which could recognize tumor cells, lymphocytes, and stromal cells autonomously and accurately within the images. This study introduces a novel immune classification system based on the modified CNN model.MethodPatients with HCC from both Beijing Hospital and The Cancer Genome Atlas (TCGA) database were included in this study. Additionally, least absolute shrinkage and selection operator (LASSO) analyses, along with logistic regression, were utilized to develop a prognostic model. We proposed an immune classification based on the percentage of lymphocytes, with a threshold set at the median lymphocyte percentage.ResultPatients were categorized into high or low infiltration subtypes based on whether their lymphocyte percentages were above or below the median, respectively. Patients with different immune infiltration subtypes exhibited varying clinical features and distinct TME characteristics. The low-infiltration subtype showed a higher incidence of hypertension and fatty liver, more advanced tumor stages, downregulated immune-related genes, and higher infiltration of immunosuppressive cells. A reliable prognostic model for predicting early recurrence of HCC based on clinical features and immune classification was established. The area under the curve (AUC) of the receiver operating characteristic (ROC) curves was 0.918 and 0.814 for the training and test sets, respectively.DiscussionIn conclusion, we proposed a novel immune classification system based on cell information extracted from pathological slices, provides a novel tool for prognostic evaluation in HCC.

Enhancing multimedia management: cloud-based movie type recognition with hybrid deep learning architecture

Film and movie genres play a pivotal role in captivating relevant audiences across interactive multimedia platforms. With a focus on entertainment, streaming providers are increasingly prioritizing the automatic generation of movie genres within cloud-based media services. In service management, the integration of a hybrid convolutional network proves to be instrumental in effectively distinguishing between a diverse array of video genres. This classification process not only facilitates more refined recommendations and content filtering but also enables targeted advertising. Furthermore, given the frequent amalgamation of components from various genres in cinema, there arises a need for social media networks to incorporate real-time video classification mechanisms for accurate genre identification. In this study, we propose a novel architecture leveraging deep learning techniques for the detection and classification of genres in video films. Our approach entails the utilization of a bidirectional long- and short-term memory (BiLSTM) network, augmented with video descriptors extracted from EfficientNet-B7, an ImageNet pre-trained convolutional neural network (CNN) model. By employing BiLSTM, the network acquires robust video representations and proficiently categorizes movies into multiple genres. Evaluation on the LMTD dataset demonstrates the substantial improvement in the performance of the movie genre classifier system achieved by our proposed architecture. Notably, our approach achieves both computational efficiency and precision, outperforming even the most sophisticated models. Experimental results reveal that EfficientNet-BiLSTM achieves a precision rate of 93.5%. Furthermore, our proposed architecture attains state-of-the-art performance, as evidenced by its F1 score of 0.9012.

Diagnosing the degree of differentiation between types of oral cancer based on extreme deep neural network model and Raman spectroscopy

As a highly prevalent and recurrent cancer, detecting the degree of differentiation in oral cancer is crucial. Current methods rely on biopsies in the presence of significant lesions, which are time-consuming. This study introduces an efficient and rapid approach for mass determination of oral cancer differentiation levels. By leveraging serum Raman spectroscopy combined with deep neural networks and extreme gradient boosting, we performed feature selection and interaction on oral cancer samples of varying differentiation levels, achieving the most accurate and reliable classification of oral cancer differentiation stages.

Prediction of the effect of adsorption on the retention of organic compounds by NF/RO using QSPR-ANN

&lt;p&gt;Understanding the retention of organic compounds (OCs) is critical for membrane applications in water recycling. The objective of this study was to create an optimized model using Artificial Neural Networks for Quantitative Structure-Property Relationship (QSPR-ANN) to predict the effect of adsorption on the retention of organic compounds (OCs) by nanofiltration (NF) and reverse osmosis (RO). An optimal model (QSPR-ANNoptimal) characterized by a similar structure (13 neurons in the inputs layer, 11 neurons in the hidden layer, and 1 neuron in the output layer) is constructed to predict the effect of adsorption on the retention of organic compounds by membranes. A set of 273 data points was used to test the neural network. the data set was used 70% for training, 15% for validation, and 15% for testing. For the most promising neural network model, the calculated retention values were compared to the experimental retention values, and good correlations were found (the determination coefficient "R2 = 0.9872" and the root mean squared error "RMSE = 2.2743%" for the test phase). This indicates the good robustness of the established QSPR-ANN model and the possibility of predicting the various parameters that characterize the retention of OCs by RO/NF. Sensitivity analysis revealed that the effect of adsorption retention of organic compounds by reverses osmosis and nanofiltration membranes depends more precisely on two important interactions (hydrophobic/adsorption and steric hindrance).&lt;/p&gt;

P20 Characterizing monoamine signalling in human dermal fibroblasts and keratinocytes

Abstract Introduction and aims Our mental health depends upon a fine balance of monoamines in the brain. Interestingly, there is ample evidence indicating that mental health disorders such as stress or anxiety have an impact on skin condition and appearance. For example, stress induces neurogenic inflammation, which in turn leads to mast cell degranulation in the skin and increases local histamine levels. This increased histamine within the skin exacerbates skin conditions such as eczema and psoriasis, while impacting the skin barrier. In this study, we first sought to investigate monoamine signalling in both skin keratinocytes and fibroblasts, with a specific focus on histamine and serotonin modulation. Next, we investigated mechanisms of release via pharmacological intervention to block corresponding receptors, informed by using single-cell RNA sequencing from existing datasets. Methods We adapted a powerful electrochemical technique, fast-scan cyclic voltammetry (FSCV) to codetect monoamine signalling in real-time in vitro. Employing the advanced capabilities of our artificial neural network models, we analysed extensive FSCV colour plot data in real-time to establish neurotransmitter release patterns and understand their characteristics. Results Our Results show that both keratinocytes and fibroblasts are capable of releasing serotonin and histamine without external intervention. After administering diphenhydramine, an H1 receptor inhibitor, to the cells, we observed a rise in the frequency and amplitude of both serotonin and histamine signalling compared with the cells that received no treatment. However, bafilomycin, a vacuolar H+ ATPase inhibitor, reduced the signalling frequency significantly. This suggests a potential autoregulation mechanism and interdependency between the two monoamines. Conclusions Overall, using analytical Methods, our work shows that skin-resident fibroblasts and keratinocytes can release monoamines including serotonin and histamine. This finding provides an opportunity to examine interventions that facilitate serotonin release, to decrease histamine levels, and improve the skin function and general mental wellbeing.

Construction and Implementation of Content-Based National Music Retrieval Model Under Deep Learning

This research mainly studies the construction and implementation of the content-based folk music retrieval model. Firstly, it studies the music automatic annotation method based on deep learning, and then proposes the tag conditional random field music automatic annotation method, and then constructs the music annotation depth neural network model combining a variety of music representation and attention mechanism. Finally, it analyzes the proposed folk music retrieval model the effectiveness of the cable model is verified and its performance is evaluated. The results show that in Glu module, Glu blocks had better performance in music annotation, and the music annotation results of each index in music hierarchical sequence modeling are better, which ensures the effectiveness of music annotation. Compared with other algorithms, the AUC tag score of the proposed method is the highest, which is 0.913; it can better model the mapping relationship between the audio features of music input to the text tag and has higher scores on all evaluation indicators.

The Impact of Short Video Live Broadcast on the Sales of Sports Machinery and Equipment

This paper proposes a neural network model based on deep learning, which can better analyze the impact of short video live broadcast on sports machinery and equipment. Firstly, this paper proposes a U-Net-based convolutional neural network as the backbone network of this paper, which mainly realizes the impact of short video live broadcast on sales. Secondly, this paper proposes a dense residual module based on the Transformer lightweight module, which can effectively improve the global modeling ability of the network model and improve the prediction accuracy of the network model. Finally, through a large number of experiments, it is proved that the convolutional neural network based on U-Net proposed in this paper can be better used for the task of short video live broadcast for the sales of sports machinery and equipment, and achieves better prediction accuracy and reasoning speed.

Urolithin A Inhibits Anterior Basolateral Amygdala to Ventral Hippocampal CA1 Circuit to Ameliorate Amyloid-β-Impaired Social Ability.

Anxiety and social withdrawal are highly prevalent among patients with Alzheimer's disease (AD). However, the neural circuit mechanisms underlying these symptoms remain elusive, and there is a need for effective prevention strategies. This study aims to elucidate the neural circuitry mechanisms underlying social anxiety in AD. We utilized 5xFAD mice and conducted a series of experiments including optogenetic manipulation, Tandem Mass Tag-labeled proteome analysis, behavioral assessments, and immunofluorescence staining. In 5xFAD mice, we observed significant amyloid-β (Aβ) accumulation in the anterior part of basolateral amygdala (aBLA). Behaviorally, 6-month-old 5xFAD mice displayed excessive social avoidance during social interaction. Concurrently, the pathway from aBLA to ventral hippocampal CA1 (vCA1) was significantly activated and exhibited a disorganized firing patterns during social interaction. By optogenetically inhibiting the aBLA-vCA1 pathway, we effectively improved the social ability of 5xFAD mice. In the presence of Aβ accumulation, we identified distinct changes in the protein network within the aBLA. Following one month of administration of Urolithin A (UA), we observed significant restoration of the abnormal protein network within the aBLA. UA treatment also attenuated the disorganized firings of the aBLA-vCA1 pathway, leading to an improvement in social ability. The aBLA-vCA1 circuit is a vulnerable pathway in response to Aβ accumulation during the progression of AD and plays a crucial role in Aβ-induced social anxiety. Targeting the aBLA-vCA1 circuit and UA administration are both effective strategies for improving the Aβ-impaired social ability.

Forecasting Next-Time-Step Forex Market Stock Prices Using Neural Networks

Purpose: This study aims to predict the closing price of the EUR/JPY currency pair in the forex market using recurrent neural network (RNN) architectures, namely Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), with the incorporation of Bidirectional layers. Methods: The dataset comprises hourly price data obtained from Yahoo Finance and pre-processed accordingly. The data is divided into training and testing sets, and time series sequences are constructed for input into the models. The RNN, LSTM, and GRU models are trained using the Adam optimization algorithm with the mean squared error (MSE) loss metric. Results: Results indicate that the LSTM model, particularly when coupled with Bidirectional layers, exhibits superior predictive performance compared to the other models, as evidenced by lower MSE values. Conclusions: Therefore, the LSTM model with Bidirectional layers is the most effective in predicting the EUR/JPY currency pair's closing price in the forex market. These findings offer valuable insights for practitioners and researchers involved in financial market prediction and neural network modelling

A Multiple Controlled Toffoli Driven Adaptive Quantum Neural Network Model for Dynamic Workload Prediction in Cloud Environments.

The key challenges in cloud computing encompass dynamic resource scaling, load balancing, and power consumption. Accurate workload prediction is identified as a crucial strategy to address these challenges. Despite numerous methods proposed to tackle this issue, existing approaches fall short of capturing the high-variance nature of volatile and dynamic cloud workloads. Consequently, this paper introduces a novel model aimed at addressing this limitation. This paper presents a novel Multiple Controlled Toffoli-driven Adaptive Quantum Neural Network (MCT-AQNN) model to establish an empirical solution to complex, elastic as well as challenging workload prediction problems by optimizing the exploration, adaption, and exploitation proficiencies through quantum learning. The computational adaptability of quantum computing is ingrained with machine learning algorithms to derive more precise correlations from dynamic and complex workloads. The furnished input data point and hatched neural weights are refitted in the form of qubits while the controlling effects of Multiple Controlled Toffoli (MCT) gates are operated at the hidden and output layers of Quantum Neural Network (QNN) for enhancing learning capabilities. Complimentarily, a Uniformly Adaptive Quantum Machine Learning (UAQL) algorithm has evolved to functionally and effectually train the QNN. The extensive experiments are conducted and the comparisons are performed with state-of-the-art methods using four real-world benchmark datasets. Experimental results evince that MCT-AQNN has up to 32%-96% higher accuracy than the existing approaches.

Development of a neural network model for peeling-ballooning stability analysis in the KSTAR tokamak pedestals

Abstract The neural network model, MISHKA-NN is developed to mitigate the computational burden associated with the linear ideal MHD stability analysis of the pedestal based on the peeling-ballooning model. By utilizing both 1D plasma profiles (current density, pressure gradient, and safety factor) and 0D parameters (plasma geometry, total current, and toroidal mode number), the model predicts linear growth rate of edge-localized ideal MHD instability in a given equilibrium state. By enabling the prediction of each instability within a second, the model reduces the time required for plotting a pedestal peeling-ballooning stability diagram ($j-\alpha$ diagram) from approximately 100 CPU hours to a few CPU minutes. Notably, even with the utilization of parametric pressure and current profiles and plasma boundary shapes for the training dataset, the model shows a satisfactory level of performance in benchmarking the $j-\alpha$ diagram for the reconstructed equilibrium from a KSTAR tokamak experiment. We anticipate the model to serve as a versatile alternative to 2D linear MHD stability codes, alleviating numerical costs.

Establishment and validation of an artificial intelligence web application for predicting postoperative in-hospital mortality in patients with hip fracture: a National cohort study of 52,707 cases.

In-hospital mortality following hip fractures is a significant concern, and accurate prediction of this outcome is crucial for appropriate clinical management. Nonetheless, there is a lack of effective prediction tools in clinical practice. By utilizing artificial intelligence and machine learning techniques, this study aims to develop a predictive model that can assist clinicians in identifying geriatric hip fracture patients at a higher risk of in-hospital mortality. A total of 52,707 geriatric hip fracture patients treated with surgery from 90 hospitals were included in this study. The primary outcome was postoperative in-hospital mortality. The patients were randomly divided into two groups, with a ratio of 7:3. The majority of patients, assigned to the training cohort, were used to develop the AI models. The remaining patients, assigned to the validation cohort, were used to validate the models. Various machine learning algorithms, including logistic regression (LR), decision tree (DT), naïve Bayesian (NB), neural network (NN), eXGBoosting machine (eXGBM), and random forest (RF), were employed for model development. A comprehensive scoring system, incorporating 10 evaluation metrics, was developed to assess the prediction performance, with higher scores indicating superior predictive capability. Based on the best machine learning-based model, an AI application was developed on the Internet. In addition, a comparative testing of prediction performance between doctors and the AI application. The eXGBM model exhibited the best prediction performance, with an AUC of 0.908 (95% CI: 0.881-0.932), as well as the highest accuracy (0.820), precision (0.817), specificity (0.814), and F1 score (0.822), and the lowest Brier score (0.120) and log loss (0.374). Additionally, the model showed favorable calibration, with a slope of 0.999 and an intercept of 0.028. According to the scoring system incorporating 10 evaluation metrics, the eXGBM model achieved the highest score (56), followed by the RF model (48) and NN model (41). The LR, DT, and NB models had total scores of 27, 30, and 13, respectively. The AI application has been deployed online at https://in-hospitaldeathinhipfracture-l9vhqo3l55fy8dkdvuskvu.streamlit.app/, based on the eXGBM model. The comparative testing revealed that the AI application's predictive capabilities significantly outperformed those of the doctors in terms of AUC values (0.908 vs. 0.682, P<0.001). The eXGBM model demonstrates promising predictive performance in assessing the risk of postoperative in-hospital mortality among geriatric hip fracture patients. The developed AI model serves as a valuable tool to enhance clinical decision-making.

Improving GOCI ocean color data under high solar-zenith angle over open oceans using neural networks

ABSTRACT With hourly measurements available during daytime between local times of 09:00–16:00, ocean color data derived from the Geostationary Ocean Color Imager (GOCI) onboard the Korean Communication, Ocean, and Meteorological (COMS) satellite have been useful for research and surveillance of diurnal processes in the western Pacific Ocean region. However, in early morning and late afternoon measurements, there are significant errors in GOCI-derived ocean color products as the solar-zenith angle (θ 0 ) goes beyond 70°, especially in autumn and winter seasons. In this study, we employ a neural network (NN) model to make corrections on the GOCI-measured normalize water-leaving radiance spectra, nL w (λ), with high θ 0 (>70°) in open oceans. Results show that NN-corrected nL w (λ) are consistent with the previous-hour nL w (λ) and make the diurnal variations in the region much more stable and reasonable. Specifically, the GOCI-measured nL w (λ) with θ 0 ≤65° in earlier hours of the day (including some nL w (λ) diurnal variation) are considered accurate and used as the ground truth to train NN models for nL w (λ) correction with high θ 0 . Further analysis of the relationship of ratios (between the NN-corrected and original nL w (λ)) with θ 0 shows that the nL w (λ) ratios increase as θ 0 increase, which indicates that there are more significant corrections with larger θ 0 (>70°). The performance evaluation of the NN models is based on the comparison of NN-corrected nL w (λ) with the original previous hour nL w (λ) data. The ratios of NN-corrected nL w (λ) to the original previous hour nL w (λ) are 0.968–1.045 for the short blue/blue and green bands, and the performance of nL w (λ) correction at 14:00 and 15:00 is slightly better than that at 16:00, due to significantly large θ 0 at late afternoon hours. The NN-corrected nL w (λ) data are also used to derive chlorophyll-a (Chl-a) concentration, showing significantly improved Chl-a in GOCI’s late afternoon measurements with θ 0 >70°.

1D-CNN-based audio tampering detection using ENF signals

The extensive adoption of digital audio recording has revolutionized its application in digital forensics, particularly in civil litigation and criminal prosecution. Electric network frequency (ENF) has emerged as a reliable technique in the field of audio forensics. However, the absence of comprehensive ENF reference datasets limits current ENF-based methods. To address this, this study introduces ATD, a blind audio forensics framework based on a one-dimensional convolutional neural network (1D-CNN) model. ATD can identify phase mutations and waveform discontinuities within the tampered ENF signal, without relying on an ENF reference database. To enhance feature extraction, the framework incorporates characteristics of the fundamental harmonics of ENF signals. In addition, a denoising method termed ENF noise reduction (ENR) based on the variational mode decomposition (VMD) and robust filtering algorithm (RFA) is proposed to reduce the impact of external noise on embedded electric network frequency signals. This study investigates three distinct types of audio tampering—deletion, insertion, and replacement—culminating in the design of binary-class tampering detection scenarios and four-class tampering detection scenarios tailored to these tampering types. ATD achieves a tampering detection accuracy of over 93% in the four-class scenario and exceeds 96% in the binary-class scenario. The effectiveness, efficiency, adaptability, and robustness of ATD in the two and four classification scenarios have been confirmed by extensive experiments.

Deep Learning‐Driven Robust Glucose Sensing and Fruit Brix Estimation Using a Single Microwave Split Ring Resonator

AbstractExtracting the desired information from sensor data with various internal and external effects is a significant challenge in sensor applications. Difficult‐to‐control factors such as temperature, humidity, and sample position can significantly affect the stability and reliability of sensor data. In this paper, a deep learning‐based glucose sensing method that is robust to variations in sample position is proposed. It is shown that the variations in sample position affect the sensor data measured by the designed split ring resonator‐based microwave sensor. Then, artificial neural network and 1D convolutional neural network (CNN) models are evaluated for extracting glucose concentration information from the sensor data measured at random sample positions. The concentration of the glucose solution ranged from 1% to 23% (2% increments). The 1D CNN with all frequencies (0.5–18 GHz) of the and datasets outperformed the other model, with a mean absolute error (MAE) of 0.695% and a mean squared error (MSE) of 0.876 evaluated via cross‐validation. The study demonstrated that the sensor system can be applied in real life by performing fruit Brix estimation based on transfer learning of the previous 1D CNN network, and the MAE and MSE are 0.450% and 0.305, respectively.

Artificial Neural Network-Based Modelling for Yield Strength Prediction of Austenitic Stainless-Steel Welds

This study aimed to develop an artificial neural network (ANN) model for predicting the yield strength of a weld metal composed of austenitic stainless steel and compare its performance with that of conventional multiple regression and machine learning models. The input parameters included the chemical composition of the nine effective elements (C, Si, Mn, P, S, Ni, Cr, Mo, and Cu) and the heat input per unit length. The ANN model (comprising five nodes in one hidden layer), which was constructed and trained using 60 data points, yielded an R2 value of 0.94 and a mean average percent error (MAPE) of 2.29%. During model verification, the ANN model exhibited superior prediction performance compared with the multiple regression and machine learning models, achieving an R2 value of 0.8644 and a MAPE of 3.06%. Consequently, the ANN model effectively predicted the variation in the yield strength and microstructure resulting from the thermal history and dilution during the welding of 3.5–9% Ni steels with stainless steel-based welding consumables. Furthermore, the application of the prediction model was demonstrated in the design of welding consumables and heat input for 9% Ni steel.

Performance Evaluation of Convolutional Neural Network Models for Classification of Highway Hidden Distresses with GPR B-Scan Images

Despite the considerable advancements in automated identification methods of highway hidden distress with ground-penetrating radar (GPR) images, there still exist challenges in realizing automated identification of highway hidden distress owing to the quantity, variability, and reliability of the distress samples and diversity of classification models. Firstly, the dataset collected contains 31,640 samples categorized into four categories: interlayer debonding, interlayer loosening, interlayer water seepage, and structural loosening from 1500 km highway, for obtaining larger enough samples and covering the variable range of distress samples. Secondly, the distresses were labeled by experienced experts, and the labels were verified with drilled cores to ensure their reliability. Lastly, 18 exemplary convolutional neural network (CNN) models from 8 different architectures were evaluated using evaluation metrics such as precision, recall, and f1-score. Further, confusion matrix and Grad-CAM techniques were utilized to analyze these models. The experimental results show that VGG13 performed most prominently and stably, while the lightweight network SqueezeNet1_1 performed particularly well with a batch size of 64. Furthermore, this study indicates that models with fewer layers can achieve comparable or better performance than deeper models.