Hybrid Deep Neural Network Research Articles

MO-0959 A novel multi-task hybrid deep neural network (DNN) predicts tumor progression during MRgRT

MO-0959 A novel multi-task hybrid deep neural network (DNN) predicts tumor progression during MRgRT

Short-term regional ionospheric TEC forecast using a hybrid deep learning neural network

The total electron content (TEC) of the ionosphere is one of the key parameters for both scientific study and engineering application. A hybrid deep learning network based on Grapy Convolution Network (GCN) and Long Short Term Memory (LSTM) is proposed to predict the regional ionospheric TEC 24 h in advance across China. The extensive 14-year dataset during 2005–2018 from the CODE TEC measurements, covering a wide variety of solar activity, are used to comprehensively evaluate the performance of the predictive model. The experimental results show that the model predicted TEC is in good agreement with the actual CODE TEC. Over 90% of differences between the actual TEC and the predicted TEC are within ±2 TECU, and the model errors obviously depend on seasons, latitudes and solar activity. Comparison with the previous work indicate that the proposed regional forecast model is competitive.

Large-Scale Road Network Traffic Congestion Prediction Based on Recurrent High-Resolution Network

Traffic congestion is a significant problem that adversely affects the economy, environment, and public health in urban areas worldwide. One promising solution is to forecast road-level congestion levels in the short-term and long-term, enabling commuters to avoid congested areas and allowing traffic agencies to take appropriate action. In this study, we propose a hybrid deep neural network algorithm based on High-Resolution Network (HRNet) and ConvLSTM decoder for 10, 30, and 60-min traffic congestion prediction. Our model utilizes the HRNet’s multi-scale feature extraction capability to capture rich spatial features from a sequence of past traffic input images. The ConvLSTM module learns temporal information from each HRNet multi-scale output and aggregates all feature maps to generate accurate traffic forecasts. Our experiments demonstrate that the proposed model can efficiently and effectively learn both spatial and temporal relationships for traffic congestion and outperforms four other state-of-the-art architectures (PredNet, UNet, ConvLSTM, and Autoencoder) in terms of accuracy, precision, and recall. A case study was conducted on the dataset from Seoul, South Korea.

Search for mono-Higgs signals in bb¯ final states using deep neural networks

We study mono-Higgs signatures emerging in an illustrative new physics scenario involving Standard Model Higgs boson decays to bottom quark pairs using hybrid deep neural networks. We use a multilayer perceptron to analyze the kinematic observables and optimize the signal-to-background discrimination. The global color flow structure of hard jets emerging from the decay of heavy particles with different color charges is crucial to single out the mono-Higgs signature. Upon embedding the different color flow structures for signal and backgrounds into constructed images, we use a convolution neural network to analyze the latter. Specifically, the approach takes initially mono-type data as input, frittering away invaluable multisource and multiscale information. We then discuss a general architecture of hybrid deep neural networks that supports instead mixed input data. In comparison with single input deep neural networks, like multilayer perceptron or convolution neural networks, the hybrid deep neural networks provide higher capacity in feature extraction and thus in signal versus background classification performance. We provide reference results for the case of the high-luminosity Large Hadron Collider.

A Hybrid Deep Neural Network Architecture for Day-Ahead Electricity Forecasting: Post-COVID Paradigm

Predicting energy demand in adverse scenarios, such as the COVID-19 pandemic, is critical to ensure the supply of electricity and the operation of essential services in metropolitan regions. In this paper, we propose a deep learning model to predict the demand for the next day using the “IEEE DataPort Competition Day-Ahead Electricity Demand Forecasting: Post-COVID Paradigm” database. The best model uses hybrid deep neural network architecture (convolutional network–recurrent network) to extract spatial-temporal features from the input data. A preliminary analysis of the input data was performed, excluding anomalous variables. A sliding window was applied for importing the data into the network input. The input data was normalized, using a higher weight for the demand variable. The proposed model’s performance was better than the models that stood out in the competition, with a mean absolute error of 2361.84 kW. The high similarity between the actual demand curve and the predicted demand curve evidences the efficiency of the application of deep networks compared with the classical methods applied by other authors. In the pandemic scenario, the applied technique proved to be the best strategy to predict demand for the next day.

Identification of thrombopoiesis inducer based on a hybrid deep neural network model

Thrombocytopenia is a common haematological problem worldwide. Currently, there are no relatively safe and effective agents for the treatment of thrombocytopenia. To address this challenge, we propose a computational method that enables the discovery of novel drug candidates with haematopoietic activities. Based on different types of molecular representations, three deep learning (DL) algorithms, namely recurrent neural networks (RNNs), deep neural networks (DNNs), and hybrid neural networks (RNNs+DNNs), were used to develop classification models to distinguish between active and inactive compounds. The evaluation results illustrated that the hybrid DL model exhibited the best prediction performance, with an accuracy of 97.8 % and Matthews correlation coefficient of 0.958 on the test dataset. Subsequently, we performed drug discovery screening based on the hybrid DL model and identified a compound from the FDA-approved drug library that was structurally divergent from conventional drugs and showed a potential therapeutic action against thrombocytopenia. The novel drug candidate wedelolactone significantly promoted megakaryocyte differentiation in vitro and increased platelet levels and megakaryocyte differentiation in irradiated mice with no systemic toxicity. Overall, our work demonstrates how artificial intelligence can be used to discover novel drugs against thrombocytopenia.

Parkinson’s disease detection and classification using EEG based on deep CNN-LSTM model

ABSTRACT The progressive loss of motor function in the brain is a hallmark of Parkinson’s disease (PD). Electroencephalogram (EEG) signals are commonly used for early diagnosis since they are associated with a brain disorder. This work aims to find a better way to represent electroencephalography (EEG) signals and enhance the classification accuracy of individuals with Parkinson’s disease using EEG signals. In this paper, we present two hybrid deep neural networks (DNN) that combine convolutional neural networks with long short-term memory to diagnose Parkinson’s disease using EEG signals, that is, through the establishment of parallel and series combined models. The deep CNN network is utilized to acquire the structural features of ECG signals and extract meaningful information from them, after which the signals are sent via a long short-term memory network to extract the features’ context dependency. The proposed architecture was able to achieve 97.6% specificity, 97.1% sensitivity, and 98.6% accuracy for a parallel model and 99.1% specificity, 98.5% sensitivity, and 99.7% accuracy for a series model, both in 3-class classification (PD patients with medication, PD patients without medication and healthy).

Automated-tuned hyper-parameter deep neural network by using arithmetic optimization algorithm for Lorenz chaotic system

<p>Deep neural networks (DNNs) are very dependent on their parameterization and require experts to determine which method to implement and modify the hyper-parameters value. This study proposes an automated-tuned hyper-parameter for DNN using a metaheuristic optimization algorithm, arithmetic optimization algorithm (AOA). AOA makes use of the distribution properties of mathematics’ primary arithmetic operators, including multiplication, division, addition, and subtraction. AOA is mathematically modeled and implemented to optimize processes across a broad range of search spaces. The performance of AOA is evaluated against 29 benchmark functions, and several real-world engineering design problems are to demonstrate AOA’s applicability. The hyper-parameter tuning framework consists of a set of Lorenz chaotic system datasets, hybrid DNN architecture, and AOA that works automatically. As a result, AOA produced the highest accuracy in the test dataset with a combination of optimized hyper-parameters for DNN architecture. The boxplot analysis also produced the ten AOA particles that are the most accurately chosen. Hence, AOA with ten particles had the smallest size of boxplot for all hyper-parameters, which concluded the best solution. In particular, the result for the proposed system is outperformed compared to the architecture tested with particle swarm optimization.</p>

Finnish parliament ASR corpus: Analysis, benchmarks and statistics.

Public sources like parliament meeting recordings and transcripts provide ever-growing material for the training and evaluation of automatic speech recognition (ASR) systems. In this paper, we publish and analyse the Finnish Parliament ASR Corpus, the most extensive publicly available collection of manually transcribed speech data for Finnish with over 3000h of speech and 449 speakers for which it provides rich demographic metadata. This corpus builds on earlier initial work, and as a result the corpus has a natural split into two training subsets from two periods of time. Similarly, there are two official, corrected test sets covering different times, setting an ASR task with longitudinal distribution-shift characteristics. An official development set is also provided. We developed a complete Kaldi-based data preparation pipeline and ASR recipes for hidden Markov models (HMM), hybrid deep neural networks (HMM-DNN), and attention-based encoder-decoders (AED). For HMM-DNN systems, we provide results with time-delay neural networks (TDNN) as well as state-of-the-art wav2vec 2.0 pretrained acoustic models. We set benchmarks on the official test sets and multiple other recently used test sets. Both temporal corpus subsets are already large, and we observe that beyond their scale, HMM-TDNN ASR performance on the official test sets has reached a plateau. In contrast, other domains and larger wav2vec 2.0 models benefit from added data. The HMM-DNN and AED approaches are compared in a carefully matched equal data setting, with the HMM-DNN system consistently performing better. Finally, the variation of the ASR accuracy is compared between the speaker categories available in the parliament metadata to detect potential biases based on factors such as gender, age, and education.

Developing digital twin design for enhanced productivity of an automated anodizing industry and process prediction using hybrid deep neural network

Automation is beneficial when implemented in challenging environments requiring less human effort or reducing human effort. Employee safety is a concern to increase productivity, delivery and achieve the required quality. The present research investigates the possibility of installing an automation procedure in an anodizing industry called GOLDEN ANODIZER, which is situated in Coimbatore, India. During the preliminary analysis at the specified industry premises, there was a reduction in productivity, quality and customer delivery due to manual operations, including some health issues were identified. And hence the present investigation analyses the possibility of installation of an automation system for the anodizing process. For such reason, the digital twin of an automation system is first designed, developed and tested using Siemens NX software. The anodizing factors such as Anodizing medium temperature (K), Acid Concentration (wt%), applied voltage (V) and Responses Surface Finish (Ra), Film Thickness (tf) and Time Duration (T) were optimized for improved productivity and quality. Prediction results and RMSE analysis show that the proposed hybrid PSO-LFA algorithm has outperformed all modern algorithms. The proposed algorithm optimized the anodizing parameters and suggested 3650.7 s as a new cycle time. Also, the improved cycle time can boost the plant outcome by 182% and reduce the manpower by 46% through the proposed Automation system.

Geometric prior guided hybrid deep neural network for facial beauty analysis

AbstractFacial beauty analysis is an important topic in human society. It may be used as a guidance for face beautification applications such as cosmetic surgery. Deep neural networks (DNNs) have recently been adopted for facial beauty analysis and have achieved remarkable performance. However, most existing DNN‐based models regard facial beauty analysis as a normal classification task. They ignore important prior knowledge in traditional machine learning models which illustrate the significant contribution of the geometric features in facial beauty analysis. To be specific, landmarks of the whole face and facial organs are introduced to extract geometric features to make the decision. Inspired by this, we introduce a novel dual‐branch network for facial beauty analysis: one branch takes the Swin Transformer as the backbone to model the full face and global patterns, and another branch focuses on the masked facial organs with the residual network to model the local patterns of certain facial parts. Additionally, the designed multi‐scale feature fusion module can further facilitate our network to learn complementary semantic information between the two branches. In model optimisation, we propose a hybrid loss function, where especially geometric regulation is introduced by regressing the facial landmarks and it can force the extracted features to convey facial geometric features. Experiments performed on the SCUT‐FBP5500 dataset and the SCUT‐FBP dataset demonstrate that our model outperforms the state‐of‐the‐art convolutional neural networks models, which proves the effectiveness of the proposed geometric regularisation and dual‐branch structure with the hybrid network. To the best of our knowledge, this is the first study to introduce a Vision Transformer into the facial beauty analysis task.

DeepFormer: a hybrid network based on convolutional neural network and flow-attention mechanism for identifying the function of DNA sequences.

Identifying the function of DNA sequences accurately is an essential and challenging task in the genomic field. Until now, deep learning has been widely used in the functional analysis of DNA sequences, including DeepSEA, DanQ, DeepATT and TBiNet. However, these methods have the problems of high computational complexity and not fully considering the distant interactions among chromatin features, thus affecting the prediction accuracy. In this work, we propose a hybrid deep neural network model, called DeepFormer, based on convolutional neural network (CNN) and flow-attention mechanism for DNA sequence function prediction. In DeepFormer, the CNN is used to capture the local features of DNA sequences as well as important motifs. Based on the conservation law of flow network, the flow-attention mechanism can capture more distal interactions among sequence features with linear time complexity. We compare DeepFormer with the above four kinds of classical methods using the commonly used dataset of 919 chromatin features of nearly 4.9 million noncoding DNA sequences. Experimental results show that DeepFormer significantly outperforms four kinds of methods, with an average recall rate at least 7.058% higher than other methods. Furthermore, we confirmed the effectiveness of DeepFormer in capturing functional variation using Alzheimer's disease, pathogenic mutations in alpha-thalassemia and modification in CCCTC-binding factor (CTCF) activity. We further predicted the maize chromatin accessibility of five tissues and validated the generalization of DeepFormer. The average recall rate of DeepFormer exceeds the classical methods by at least 1.54%, demonstrating strong robustness.

S-Wave Velocity Forecasting Using Drill Cuttings and Deep Hybrid Neural Networks: A Case Study on a Tight Glutenite Reservoir in Mahu Sag, Junggar Basin

S-wave velocity (Vs) is a critical petrophysical parameter for reservoir characterization. It is desirable to predict Vs based on conventional logging data, but the logging cost is high. Therefore, a deep hybrid neural network coupling the convolutional neural network (CNN), Stacked gated recurrent unit (SGRU) is proposed to predict the Vs, where the inputs to the model are drill cutting features. In the proposed CNN-SGRU hybrid model, CNN is adopted to capture the spatial features from the input data, and SGRU is used to extract the temporal patterns of variation from both the forward and backward directions. To illustrate the prediction effect, the glutenite reservoir in the Baikouquan Formation of Mahu Sag, Junggar Basin is taken as an example. Mineral and pore information of drill cuttings, including siliciclastic content, clay content, quartz content, and void area ratio is chosen as the input data of the CNN-SGRU hybrid model. Three indices are used to quantitatively evaluate the prediction performance, including Mean absolute percentage error (MAPE), Root mean square error (RMSE), and Mean absolute error (MAE). The results show that the prediction accuracy of the proposed model is higher than that of the Xu-White model, CNN, and GRU. Furthermore, the results indicate that drill cuttings can replace logging data to predict Vs.

Prediction of tear propagation path of stratospheric airship envelope material based on deep learning

Tear propagation of the envelope material could cause fatal damage to the stratospheric airship (SSA) and it is very important to detect the crack and predict its tear propagation path. A hybrid deep neural network (DNN) model is proposed in this paper to predict tear propagation of the SSA envelope material mainly including stress field predictor and crack map predictor by considering the time and spatial characteristics. The Gated Recurrent Unit (GRU) is applied by using the gating network signaling that control how the present input and previous memory for the stress field predictor. A Feature Pyramid Network (FPN)-based faster region-based Convolutional Neural Network (CNN) is proposed to predict the crack location by declaring the crack propagation direction and velocity of the envelope material. Furthermore, two deformable operation modules are embedded into the crack detector to achieve better identification of out-of-plane cracks of the envelope material for a real airship with curvature. The dataset is obtained by extended finite element method (XFEM) analysis. The proposed approach has potential applications in the field of envelope material design and structural health monitoring of the SSA.

A hybrid discriminant fuzzy DNN with enhanced modularity bat algorithm for speech recognition

In recent years, speech processing resides a major application in the domain of signal processing. Due to the audibility loss of some speech signals, people with hearing impairment have difficulty in understanding speech, which reintroduces a crucial role in speech recognition. Automatic Speech Recognition (ASR) development is a major challenge in research in the case of noise, domain, vocabulary size, and language and speaker variability. Speech recognition system design needs careful attention to challenges or issues like performance and database evaluation, feature extraction methods, speech representations and speech classes. In this paper, HDF-DNN model has been proposed with the hybridization of discriminant fuzzy function and deep neural network for speech recognition. Initially, the speech signals are pre-processed to eliminate the unwanted noise and the features are extracted using Mel Frequency Cepstral Coefficient (MFCC). A hybrid Deep Neural Network and Discriminant Fuzzy Logic is used for assisting hearing-impaired listeners with enhanced speech intelligibility. Both DNN and DF have some problems with parameters to address this problem, Enhanced Modularity function-based Bat Algorithm (EMBA) is used as a powerful optimization tool. The experimental results show that the proposed automatic speech recognition-based hybrid deep learning model is effectively-identifies speech recognition more than the MFCC-CNN, CSVM and Deep auto encoder techniques. The proposed method improves the overall accuracy of 8.31%, 9.71% and 10.25% better than, MFCC-CNN, CSVM and Deep auto encoder respectively.

FERNET: An Integrated Hybrid DCNN Model for Driver Stress Monitoring via Facial Expressions

Drivers undergo a lot of stress that might cause distraction and might lead to an unfortunate incident. Emotional recognition via facial expressions is one of the most important field in the human–machine interface. The goal of this paper is to analyze the drivers’ facial expressions in order to monitor their stress levels. In this paper, we propose FERNET — a hybrid deep convolutional neural network model for driver stress recognition through facial emotion recognition. FERNET is an integration of two DCNNs, pre-trained ResNet101V2 CNN and a custom CNN, ConvNet4. The experiments were carried out on the widely used public datasets CK[Formula: see text], FER2013 and AffectNet, achieving the accuracies of 99.70%, 74.86% and 70.46%, respectively, for facial emotion recognition. These results outperform the recent state-of-the-art methods. Furthermore, since a few specific isolated emotions lead to higher stress levels, we analyze the results for stress- and nonstress-related emotions for each individual dataset. FERNET achieves stress prediction accuracies of 98.17%, 90.16% and 84.49% for CK[Formula: see text], FER2013 and AffectNet datasets, respectively.

Improving generalisation and accuracy of on-line milling chatter detection via a novel hybrid deep convolutional neural network

Unstable chatter seriously reduces the quality of machined workpiece and machining efficiency. In order to improve productivity, on-line chatter detection has attracted much interest in the past decades. Nevertheless, traditional methods are inevitably flawed due to the manually extracted features. Deep learning methods possess outstanding feature learning and classification capabilities, but the generalisation and accuracy are severely affected by the labelling and training of data. To address this, this paper proposed a novel hybrid deep convolutional neural network method combining an Inception module and a Squeeze-and-Excitation ResNet block (SR-block), namely ISR-CNN. The Inception module can automatically extract multi-scale features of cutting force signal to enrich the feature map. The SR-block can assign weights to different feature channels, thus suppressing useless feature maps and improving the model accuracy. Meanwhile, the introduction of SR-block also reduces the risk of gradient disappearance and speeds up the training of network. The generalisation and accuracy of the model is guaranteed by combining the two modules without training with transition state data. Milling tests were carried out on a wedge-shaped workpiece using different cutting parameters and tool overhang lengths to verify the accuracy and generalisability of the proposed method. The results showed that the proposed method outperforms other methods by achieving classification accuracy of on the validation and test sets 100% and 97.8%, respectively. In comparison to existing methods, the proposed method can correctly identify each machining state, including the transition states. Furthermore, the proposed method identifies the onset of chatter earlier than other methods, which is beneficial for chatter suppression.

Intelligent hybrid hand gesture recognition system using deep recurrent neural network with chaos game optimisation

ABSTRACT Hand gesture recognition is considered an essential task in various human-computer interaction (HCI) applications. Therefore, developing a robust system for hand gesture recognition is a challenge. This work proposed a new hybrid approach named hybrid deep recurrent neural network (RNN) incorporated with a chaos game optimisation (CGO) algorithm for efficiently recognising hand gestures. The main objective of a hybrid recurrent neural network with chaos game optimisation (RNN-CGO) is to achieve the recognition of alphabet signs from 2D gesture images. The different stages included in this technique are pre-processing, feature extraction, feature selection, and classification. The proposed approach is executed using American Sign Language (ASL) dataset with the help of the PYTHON platform. The precision, accuracy, f1-score, and recall obtained for the proposed approach are 99.28%, 99.96%, 99.25%, and 99.28%, respectively. The inference time required for the proposed approach is 0.121 s. The proposed technique has less computational complexity. The results showed that the proposed approach is better than the existing hand gesture recognition model approaches.

Traffic management in vehicular adhoc networks using hybrid deep neural networks and mobile agents

<p>The traffic congestion in vehicular adhoc networks (VANETs) is a vital problem due to its dynamic increase in traffic loads. VANETs undergo inefficient routing capability due to its increasing traffic demands. This has led to the need for intelligent transport system (ITS) to assist VANETs in enabling suitable traffic loads between vehicles and road side units (RSU). Most conventional systems offer distributed solution to manage traffic congestion but fail to regulate real-time traffic flows. In this paper, a dynamic traffic control in VANETs is offered by combining deep neural network (DNN) with mobile agents (MA). An experimental analysis is carried out to test the efficacy of the DNN-MA against conventional machine learning and a deep learning routing algorithm in VANETs. DNN-Mal is validated under various traffic congestion metrics like latency, percentage delivery ratio, packet error rate, and throughput. The results show that the proposed method offers reduced energy consumption and latency.</p>

A Hybrid CNN-LSTM Model for Predicting Solar Cycle 25

The solar cycle is linked to the number of sunspots and follows the fluctuations of the Sun’s magnetic field. It can have powerful global impacts on the Earth. Thus, predicting the timing and amplitude of the peak of the incoming solar cycle 25 is of great importance. This study uses a hybrid deep learning convolutional neural network (CNN) - long short-term memory (LSTM) model and the observed 13-month smoothed sunspot numbers to predict Solar Cycle 25. Here it is shown for the first time that the MinMax normalization method substantially reduces the error of the CNN-LSTM model’s solar cycle predictions compared to the Standard Deviation normalization method. The results also suggest that it is best to use four historical solar cycles to predict the future solar cycle. The predicted Solar Cycle 25 has a 13-month smoothed peak amplitude similar to that of Solar Cycle 24. The predicted Solar Cycle 25 peak spans a relatively long period of time between approximately August 2023 and July 2024.