Hybrid Convolutional Neural Network‐Analytical Model for Prediction of Line Edge Roughness‐Induced Performance Variations in Fin‐Shaped Field‐Effect Transistor Devices and SRAM

Landslide is a natural disaster that seriously affects human life and social development. In this study, the characteristics and effectiveness of convolutional neural network (CNN) and conventional machine learning (ML) methods in a landslide susceptibility assessment (LSA) are compared. Six ML methods used in this study are Adaboost, multilayer perceptron neural network (MLP-NN), random forest (RF), naive Bayes, decision tree (DT), and gradient boosting decision tree (GBDT). First, the basic knowledge and structures of the CNN and ML methods, and the steps of the LSA are introduced. Then, 11 conditioning factors in three categories in the Hongxi River Basin, Pingwu County, Mianyang City, Sichuan Province are chosen to build the train, validation, and test samples. The CNN and ML models are constructed based on these samples. For comparison, indicator methods, statistical methods, and landslide susceptibility maps (LSMs) are used. The result shows that the CNN can obtain the highest accuracy (86.41%) and the highest AUC (0.9249) in the LSA. The statistical methods represented by the mean and variance of TP and TN perform more firmly on the possibility of landslide occurrence. Furthermore, the LSMs show that all models can successfully identify most of the landslide points, but for areas with a low frequency of landslides, some models are insufficient. The CNN model demonstrates better results in the recognition of the landslides’ cluster region, this is also related to the convolution operation that takes the surrounding environment information into account. The higher accuracy and more concentrative possibility of CNN in LSA is of great significance for disaster prevention and mitigation, which can help the efficient use of human and material resources. Although CNN performs better than other methods, there are still some limitations, the identification of low-cluster landside areas can be enhanced by improving the CNN model.

Deep learning-based classification and mutation prediction from histopathological images of hepatocellular carcinoma.

Mark-Spectra: A convolutional neural network for quantitative spectral analysis overcoming spatial relationships

Tunnel boring machine vibration-based deep learning for the ground identification of working faces

Service-oriented model-based fault prediction and localization for service compositions testing using deep learning techniques

A device level way to quantitatively assess the influence of Line Edge Roughness (LER) on different fin based structures is considered. We know FinFETs have high drive current and offer better control over leakage and short channel effects. The FinFET device parameters such as On-current, Off-current and threshold voltage are impressionable to structure of FinFET. Performance of the 3D FinFET structures affected by LER. LER is referred to as randomly varied edges or roughness in printed pattern edge. The effect of LER in some cases results in reduction of leakage current, but in some cases it worsened the device parameter. The brunt of random variation sources on the electrical characteristics of the devices are not easy to be discriminated from each other. Random dopant fluctuations (RDF), Metal Gate Granularity (MGG) and Line Edge Roughness (LER) are the severe random variations which enforce restraints on the FinFET. Here an investigation on the effect of LER on the threshold voltage and On-current of SOI, Bulk and GAA FinFET is proposed. A comparative study on different structures is also presented. All simulations are carried out in Silvaco Atlas TCAD.

The relevance of solving the problem of facial emotion recognition on human images in the creation of modern intelligent systems of computer vision and human-machine interaction, online learning and emotional marketing, health care and forensics, machine graphics and game intelligence is shown. Successful examples of technological solutions to the problem of facial emotion recognition using transfer learning of deep convolutional neural networks are shown. But the use of such popular datasets as DISFA, CelebA, AffectNet, for deep learning of convolutional neuralnetworks does not give good results in terms of the accuracy of emotion recognition, because almost all training sets have fundamental flaws related to errors in their creation, such as the lack of data of a certain class, imbalance of classes, subjectivity and ambiguity of labeling, insufficient amount of data for deep learning, etc. It is proposed to overcome the noted shortcomings of popular datasets for emotion recognition by adding to the training sample additional pseudo-labeled images with human emotions, on which recognition occurs with high accuracy. The aim of the research is to increase the accuracy of facial emotion recognitionon the image of a human by developing a pseudo-labeling method for transfer learning of a deep neural network. To achieve the aim, the following tasks were solved: a convolutional neural network model, previously trained on the ImageNet set using the transfer learning method, was adjusted on the RAF-DB data set to solve emotion recognition tasks; a pseudo-labeling method of the RAF−DB set data was developed for semi-supervised learning of a convolutional neural network model for the task of facial emotion recognition; the accuracy of facial emotion recognition was analyzed based on the developed convolutional neural network model and the method of pseudo-labeling of RAF-DB set data for its correction. It is shown that the use of the developed method of pseudo-labeling data and transfer learning of the MobileNet V1 convolutional neural network model allowed to increase the accuracy of facial emotion recognitionon the images of the RAF-DB dataset by 2 percent (from 76 to 78%) according to the F1 estimate. Atthe same time, taking into account the significant imbalance of the classes, for the 7 main emotions in the trainingset, we have a significant increase in the accuracy of recognizing a few representatives of such emotions as surprise (from 71 to 77%), fearful(from 64 to 69%), sad (from 72 to 76%), angrywith (from 64 to 74%), neutral(from 66 to 71%). The accuracy of recognizing the emotion of happy, which is the most common, decreased (from 91 to 86 %) Thus, it can be concluded that the use of the developed pseudo-labeling method gives good results in overcoming such shortcomings of datasets for deep learning of convolutional neural networks such as lack of data of a certain type, imbalance of classes, insufficient amount of data for deep learning, etc.

Artificial intelligence: finding the intersection of predictive modeling and clinical utility

The number of traffic accidents and fatalities has increased dramatically in the modern period. Every minute, a catastrophic traffic accident occurs in India, and every hour, 16 people die on Indian roadways. The driver’s errors such as negligence, drowsiness, and wrong driving decisions lead to advancement of the autonomous vehicle industry. Such a dire situation necessitates decision-making approaches that are both computationally efficient and quick to respond. A convolutional neural network (CNN) model could be used to map the raw pixels from a single front-facing camera directly to calculate the steering commands.In this paper, a CNN methodology has been implemented for the estimation of steering angles. Here the concepts of deep learning and convolutional neural networks are applied to teach the computer to drive car autonomously. The revolutionary aspect of CNNs is that characteristics are automatically learned from training samples. Because the convolution procedure captures the 2D aspect of images. Images from the cameras (input to CNN) are fed into a CNN which then computes a proposed steering command. The command thus obtained from CNN is compared with desired command, and the CNN weights are changed accordingly to get the CNN output closer to the desired output. Back propagation algorithm is used for weight adjustments. Once trained, the network can generate steering angles from the video images of a single center camera. The data collection is performed using Udacity self-driving car simulator designed by Unity (game engine). The CNN model is able to learn meaningful road features from training signal (steering alone).

Surface roughness and machining accuracy are essential indicators of the quality of parts in milling. With recent advancements in sensor technology and data processing, the cutting force signals collected during the machining process can be used for the prediction and determination of the machining quality. Deep-learning-based artificial neural networks (ANNs) can process large sets of signal data and can make predictions according to the extracted data features. During the final stage of the milling process of SUS304 stainless steel, we selected the cutting speed, feed per tooth, axial depth of cut, and radial depth of cut as the experimental parameters to synchronously measure the cutting force signals with a sensory tool holder. The signals were preprocessed for feature extraction using a Fourier transform technique. Subsequently, three different ANNs (a deep neural network, a convolutional neural network, and a long short-term memory network) were applied for training in order to predict the machining quality under different cutting conditions. Two training methods, namely whole-data training and training by data classification, were adopted. We compared the predictive accuracy and efficiency of the training process of these three models based on the same training data. The training results and the measurements after machining indicated that in predicting the surface roughness based on the feed per tooth classification, all the models had a percentage error within 10%. However, the convolutional neural network (CNN) and long short-term memory (LSTM) models had a percentage error of 20% based on the whole-data training, while that of the deep neural network (DNN) model was over 50%. The percentage error for the machining accuracy prediction based on the whole-data training of the DNN and CNN models was below 10%, while that of the LSTM model was as large as 20%. However, there was no significant improvement in the results of the classification training. In all the training processes, the CNN model had the best analytical efficiency, followed by the LSTM model. The DNN model performed the worst.

This study proposes a method to extract the signature bands from the deep learning models of multispectral data converted from the hyperspectral data. The signature bands with two deep-learning models were further used to predict the sugar content of the Syzygium samarangense. Firstly, the hyperspectral data with the bandwidths lower than 2.5 nm were converted to the spectral data with multiple bandwidths higher than 2.5 nm to simulate the multispectral data. The convolution neural network (CNN) and the feedforward neural network (FNN) used these spectral data to predict the sugar content of the Syzygium samarangense and obtained the lowest mean absolute error (MAE) of 0.400° Brix and 0.408° Brix, respectively. Secondly, the absolute mean of the integrated gradient method was used to extract multiple signature bands from the CNN and FNN models for sugariness prediction. A total of thirty sets of six signature bands were selected from the CNN and FNN models, which were trained by using the spectral data with five bandwidths in the visible (VIS), visible to near-infrared (VISNIR), and visible to short-waved infrared (VISWIR) wavelengths ranging from 400 to 700 nm, 400 to 1000 nm, and 400 to 1700 nm. Lastly, these signature-band data were used to train the CNN and FNN models for sugar content prediction. The FNN model using VISWIR signature bands with a bandwidth of ± 12.5 nm had a minimum MAE of 0.390°Brix compared to the others. The CNN model using VISWIR signature bands with a bandwidth of ± 10 nm had the lowest MAE of 0.549° Brix compared to the other CNN models. The MAEs of the models with only six spectral bands were even better than those with tens or hundreds of spectral bands. These results reveal that six signature bands have the potential to be used in a small and compact multispectral device to predict the sugar content of the Syzygium samarangense.

In this study, we developed a novel multi-criteria decision-making (MCDM) framework for evaluating and benchmarking hybrid multi-deep transfer learning models using radiography X-ray coronavirus disease (COVID-19) images. First, we collected and pre-processed eight public databases related to the targeted datasets. Second, convolutional neural network (CNN) models extracted features from 1,338 chest X-ray (CXR) frontal view image data using six pre-trained models: VGG16, VGG19, painters, SqueezeNet, DeepLoc, and Inception v3. Then, we used the intersection between the six CNN models and eight classical machine learning (ML) methods, including AdaBoost, Decision Tree, logistic regression, random forest, kNN, neural network, and Naive Bayes, to introduce 48 hybrid classification models. In this study, eight supervised ML methods were used to classify COVID-19 CXR images. The classifiers were implemented using the TensorFlow2 and Keras libraries in Python. A feature vector was extracted from each image, and a five-fold cross-validation technique was used to evaluate the performance. The cost parameter [Formula: see text] was set to 1 and the gamma parameter [Formula: see text] was set to 0.1 for all classifiers. The experiments were run on a Windows-based computer with dual Intel I CoITM i7 processors at 2.50[Formula: see text]GHz, 8[Formula: see text]GB of RAM, and a graphical processing unit of 2[Formula: see text]GB. The performance metrics of the 48 hybrid models, including the classification accuracy (CA), specificity, area under the curve (AUC), F1 score, precision, recall, and log loss, were used as efficient evaluation criteria. Third, the MCDM approach was used, which included (i) developing a dynamic decision matrix based on seven evaluation metrics and the developed hybrid models, (ii) developing the fuzzy-weighted zero-inconsistency method for determining the weight coefficients for the seven-evaluation metrics with zero inconsistency, and (iii) developing the Višekriterijumsko Kompromisno Rangiranje method for benchmarking the 48 hybrid models. Our experimental results reveal that (i) CA and AUC obtained the highest importance weights of 0.164 and 0.147, respectively, whereas F1 and specificity obtained the lowest weights of 0.134 and 0.134, respectively, and (ii) the highest three hybrid model scores were painters neural network, painters logistic regression, and VGG16-logistic regression, making them the highest ranking scores. Finally, the developed framework was validated using sensitivity analysis and comparison analysis.

<div class="section abstract"><div class="htmlview paragraph">In the automotive embedded system domain, the measurements from vehicle and Hardware-In-Loop are currently evaluated against the testcases, either manually or via automation scripts. These evaluations are localized; they evaluate a limited number of signals for a particular measurement without considering system-level behavior. This results in defect leakage. This study aims to develop a tool that can notify anomalies at the signal level in a new measurement without referring to the testcases, considering a more significant number of system-level signals, thereby significantly reducing the defect leakage. The tool learns important features and patterns of each maneuver from many historical measurements using deep learning techniques. We tried two CNN (convolution neural network) models. The first one is a specially designed CNN that does this maneuver classification and class-specific feature extraction. The second model we tried is the FCN (Fully Convolutional Network) Classification model. CNN-based architecture can be trained faster than the recurrent neural network (RNN) model because it utilizes features extracted from the input data. A Generative Adversarial Network (GAN) model is used in series with the CNN model to clone each of these maneuvers for predicting the anomalies. During the testing phase, the CNN model maps the test measurement to the most similar maneuver from the list of already learned maneuvers, followed by the GAN model outputting the anomalies, if any. To validate the tool, 12 measurements, each of 3 different maneuvers, were selected from an old and matured function in the brake system. The class-specific feature-based classification model resulted in 33% accuracy. However, with the Fully Convolutional Network Classification model, we got 100% accuracy. We injected anomalies in one CSV file for testing purposes. The anomaly detection module predicted all the anomalies correctly. Our future goal is to implement this model at the deployment level.</div></div>

In meteorology, neural networks have the potential to be useful for advancing forecasting and prediction capabilities, especially since some were designed to be useful for time series data like weather data. This study investigated the performance of the Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN). Both CNN and RNN are ideal for time series classification problems. The models were designed to look back at 5 days of weather data to predict the presence or category of a typhoon (No Typhoon, Tropical Depression, Tropical Storm, Severe Tropical Storm, Typhoon, and Super Typhoon). The models were fed with weather data (obtained from NASA and PAGASA) from four locations in the Philippines with the parameters: atmospheric pressure, humidity, precipitation, temperature and wind speed. The research investigated the Accuracy, Cross Entropy Error, Precision, Recall, and F1-Measure, validated using 12-fold Rolling Basis Cross Validation. The results reveal that the CNN and RNN model performed to varying extents. The CNN model scored better at average accuracy however, the RNN model performed better at average cross entropy error, precision, recall, and F1 measure. The RNN model achieved better scores for precision on most categories while the CNN model performed better at recall and F1 measure on other categories. Both performed better at precision, recall and F1 measure on No Typhoon compared to other categories. This is likely due to the historical data being mostly composed of days with no typhoons.

In recent years, convolutional neural network (CNN) and other deep learning\nmodels have been gradually introduced into the area of gravitational-wave (GW)\ndata processing. Compared with the traditional matched-filtering techniques,\nCNN has significant advantages in efficiency in GW signal detection tasks. In\naddition, matched-filtering techniques are based on the template bank of the\nexisting theoretical waveform, which makes it difficult to find GW signals\nbeyond theoretical expectation. In this paper, based on the task of GW\ndetection of binary black holes, we introduce the optimization techniques of\ndeep learning, such as batch normalization and dropout, to CNN models. Detailed\nstudies of model performance are carried out. Through this study, we recommend\nto use batch normalization and dropout techniques in CNN models in GW signal\ndetection tasks. Furthermore, we investigate the generalization ability of CNN\nmodels on different parameter ranges of GW signals. We point out that CNN\nmodels are robust to the variation of the parameter range of the GW waveform.\nThis is a major advantage of deep learning models over matched-filtering\ntechniques.\n