Hybrid Neural Network Research Articles

A New Fingerprint and Graph Hybrid Neural Network for Predicting Molecular Properties.

Machine learning plays a role in accelerating drug discovery, and the design of effective machine learning models is crucial for accurately predicting molecular properties. Characterizing molecules typically involves the use of molecular fingerprints and molecular graphs. These are input into a multilayer perceptron (MLP) and variants of graph neural networks, such as graph attention networks (GATs). Due to the diverse types and large dimension of fingerprints, models may contain many features that are relatively irrelevant or redundant; meanwhile, although the GAT excels in handling heterogeneous graph tasks, it lacks the ability to extract collaborative information from neighboring nodes, which is crucial in scenarios where it cannot capture the joint influence of adjacent groups on atoms. To overcome these challenges, we introduce a hybrid model, combining improved GAT and MLP. In GAT, the recurrent neural network is employed to capture collaborative information. To address the dimensionality issue, we propose a feature selection algorithm, which is based on the principle of maximizing relevance while minimizing redundancy. Through experiments on 13 public data sets and 14 breast cell lines, our model demonstrates superior performance compared to state-of-the-art deep learning and traditional machine learning algorithms. Additionally, a series of ablation experiments were conducted to demonstrate the advantages of our improved version, as well as its antinoise capability and interpretability. These results indicate that our model holds promising prospects for practical applications.

Bidirectional long-short term memory predictor for material removal rate in computer-controlled optical surfacing

In Computer-Controlled Optical Surfacing technology, the precision of the removal function directly affects the accuracy of computer-aided processing software predictions, which in turn influences subsequent polishing machine processing. A key parameter for constructing the removal function is the material removal rate, which is often challenging to obtain accurately. Currently, the Preston equation is widely used to describe the principles of material removal. However, as a linear equation that omits many factors, it struggles to accurately model the removal function in complex machining scenarios. Therefore, this paper proposes a hybrid neural network model combining Convolutional Neural Networks and Bidirectional Long Short-Term Memory to predict the material removal rate. The model's parameters are optimized using an improved Grey Wolf Optimization algorithm, ultimately establishing a removal function closely consistent with an actual removal function. We first tested our method on the PHM2016 Data Challenge dataset, achieving a mean squared error of 6.19 and an R2 of 0.9949, outperforming other mainstream neural network prediction models developed in recent years. Additionally, we further validated the performance of the neural network using a small grinding head polishing dataset, achieving MSE and R2 values of 1.9035 and 0.99902, respectively. Finally, we applied this method to construct the removal function on an actual small grinding head production line. Compared to the traditional Preston equation-based removal function, the predicted residual surface's PV and RMS errors were reduced from 28.24 % to 35.58 %–4.563 % and 4.86 %, respectively. These validation results demonstrate that the proposed method not only facilitates easier acquisition of the removal function model but also significantly enhances the accuracy of computer-aided processing software predictions, thereby better guiding ultra-precision machining processes.

A 3-D Attention-Enhanced Hybrid Neural Network for Turbofan Engine Remaining Life Prediction Using CNN and BiLSTM Models

A 3-D Attention-Enhanced Hybrid Neural Network for Turbofan Engine Remaining Life Prediction Using CNN and BiLSTM Models

Hybrid YSGOA and neural networks based software failure prediction in cloud systems

In the realm of cloud computing, ensuring the dependability and robustness of software systems is paramount. The intricate and evolving nature of cloud infrastructures, however, presents substantial obstacles in the pre-emptive identification and rectification of software anomalies. This study introduces an innovative methodology that amalgamates hybrid optimization algorithms with Neural Networks (NN) to refine the prediction of software malfunctions. The core objective is to augment the purity metric of our method across diverse operational conditions. This is accomplished through the utilization of two distinct optimization algorithms: the Yellow Saddle Goat Fish Algorithm (YSGA), which is instrumental in the discernment of pivotal features linked to software failures, and the Grasshopper Optimization Algorithm (GOA), which further polishes the feature compilation. These features are then processed by Neural Networks (NN), capitalizing on their proficiency in deciphering intricate data patterns and interconnections. The NNs are integral to the classification of instances predicated on the ascertained features. Our evaluation, conducted using the Failure-Dataset-OpenStack database and MATLAB Software, demonstrates that the hybrid optimization strategy employed for feature selection significantly curtails complexity and expedites processing

Bee-inspired insights: Unleashing the potential of artificial bee colony optimized hybrid neural networks for enhanced groundwater level time series prediction.

Analysis of the change in groundwater used as a drinking and irrigation water source is of critical importance in terms of monitoring aquifers, planning water resources, energy production, combating climate change, and agricultural production. Therefore, it is necessary to model groundwater level (GWL) fluctuations to monitor and predict groundwater storage. Artificial intelligence-based models in water resource management have become prevalent due to their proven success in hydrological studies. This study proposed a hybrid model that combines the artificial neural network (ANN) and the artificial bee colony optimization (ABC) algorithm, along with the ensemble empirical mode decomposition (EEMD) and the local mean decomposition (LMD) techniques, to model groundwater levels in Erzurum province, Türkiye. GWL estimation results were evaluated with mean square error (MSE), coefficient of determination (R2), and residual sum of squares (RSS) and visually with violin, scatter, and time series plot. The study results indicated that the EEMD-ABC-ANN hybrid model was superior to other models in estimating GWL, with R2 values ranging from 0.91 to 0.99 and MSE values ranging from 0.004 to 0.07. It has also been revealed that promising GWL predictions can be made with previous GWL data.

Automatic Search for a Solution in the Information Management System to Monitor the State of the Production Environment

The article presents a method for monitoring and managing the state of a production environment based on an information management system with an automatic solution search module using a hybrid neural network. The object of the study is the state of the production environment. Monitoring and management are carried out on the basis of predicting changes within the environment state, taking into account the interdependence of the measured parameters. In addition to the task of regulating the value of the controlled parameter, there is an additional task of automatically compiling a control program depending on external parametersprovided by this method of monitoring and control. To do this, an automatic solution search device with software based on neural networks is installed in a closed control loop. This device analyzes many parameters of the production environment and supplies the control device with the required value of the controlled parameter. To analyze dependencies between parameters and predict the state of the production environment, a neural network is used, the weights and biases of which are calculated using the harmonic search algorithm; training was carried out with the gradient descent algorithm. To assess the accuracy and stability of the automatic search for solutions in the information management system for monitoring the state of the production environment, an experiment was conducted, the results of which made it possible to obtain the parameters of the hybrid model. Further study of the obtained parameters showed that its use in the control loop of the information system for monitoring the state of the production environment makes it possible to provide a high-quality and stable forecast of changes in the state of the production environment. The proposed method made it possible to obtain stable operation of the information and control system for monitoring the state of the production environment and maintaining controlled parameters within the acceptable range.

Hybrid neural network-aided strong wind speed prediction along rail network

With the advent of networked rail operations and increasing risks caused by extreme weather conditions, there is an urgent need to enhance the safety early warning and emergency protection for rail under extreme wind environments. By deploying anemometers along the railway and monitoring the surrounding wind speeds, dispatchers can proactively take protective measures to mitigate the impact of severe weather on rail transportation. In particular, accurate time-series forecasting during periods when wind speeds exceed the threshold is critical for railway anomaly monitoring. This study employs Extreme Value Theory (EVT) to determine the threshold for strong winds. In order to identify high wind speeds, memory networks are used to memorize strong wind events in historical data. By combining the Gated Recurrent Unit (GRU) module with the attention mechanism, this study ultimately develops a hybrid neural network model in the prediction of wind speed. Based on real-world datasets of wind speeds along the Hangzhou-Haining intercity railway and along the Norway rail network via meteorological stations, the effectiveness of the hybrid neural network model has been verified by comparing base models. The proposed strong wind speed prediction has practical applications in enhancing early warning and rail safety under complex wind climate.

One shot lumen mesh generation of abdominal aortic aneurysm by hybrid neural network

One shot lumen mesh generation of abdominal aortic aneurysm by hybrid neural network

A hybrid neural network for urban rail transit short-term flow prediction

A hybrid neural network for urban rail transit short-term flow prediction

Multi-step ahead prediction of lake water temperature using neural network and physically-based model

ABSTRACT Deep learning (DL) is a powerful tool that has proven highly effective in many applications, but creating new deep learning models is becoming increasingly challenging. However, in some fields, such as fluid dynamics, theoretical models can help design powerful DL models. Based on the existing air2water (A2W) model, this paper proposes a hybrid neural network model, DL-A2W, which combines the long short-term memory (LSTM) neural network with the A2W model to predict lake water temperature. The DL-A2W model was established using the datasets of the UK Centre for Ecology & Hydrology, and the performance of the model was evaluated through three experiments. Compared with other models, the DL-A2W model has the lowest mean absolute error (MAE), mean absolute percent error (MAPE), root mean squared error (RMSE), and the highest Nash-Sutcliffe efficiency coefficient (NSC) at any given prediction step. The values of MAE, MAPE, RMSE and NSC of the DL-A2W model on the test set were 0.223–0.388, 1.946–3.296%, 0.375–0.647 and 0.985–0.995, respectively. The results show that the DL-A2W model has good generalization ability and portability, and can accurately perform multi-step ahead prediction of lake water temperature.

Certain investigation on hybrid neural network method for classification of ECG signal with the suitable a FIR filter

The Electrocardiogram (ECG) records are crucial for predicting heart diseases and evaluating patient’s health conditions. ECG signals provide essential peak values that reflect reliable health information. Analyzing ECG signals is a fundamental technique for computerized prediction with advancements in Very Large-Scale Integration (VLSI) technology and significantly impacts in biomedical signal processing. VLSI advancements focus on high-speed circuit functionality while minimizing power consumption and area occupancy. In ECG signal denoising, digital filters like Infinite Impulse Response (IIR) and Finite Impulse Response (FIR) are commonly used. The FIR filters are preferred for their higher-order performance and stability over IIR filters, especially in real-time applications. The Modified FIR (MFIR) blocks were reconstructed using the optimized adder-multiplier block for better noise reduction performance. The MIT-BIT database is used as reference where the noises are filtered by the MFIR based on Optimized Kogge Stone Adder (OKSA). Features are extracted and analyzed using Discrete wavelet transform (DWT) and Cross Correlation (CC). At this modern era, Hybrid methods of Machine Learning (HMLM) methods are preferred because of their combined performance which is better than non-fused methods. The accuracy of the Hybrid Neural Network (HNN) model reached 92.3%, surpassing other models such as Generalized Sequential Neural Networks (GSNN), Artificial Neural Networks (ANN), Support Vector Machine with linear kernel (SVM linear), and Support Vector Machine with Radial Basis Function kernel (SVM RBF) by margins of 3.3%, 5.3%, 23.3%, and 24.3%, respectively. While the precision of the HNN is 91.1%, it was slightly lower than GSNN and ANN but higher than both SVM linear and SVM -RBF. The HNN with various features are incorporated to improve the ECG classification. The accuracy of the HNN is switched to 95.99% when the DWT and CC are combined. Also, it improvises other parameters such as precision 93.88%, recall is 0.94, F1 score is 0.88, Kappa is 0.89, kurtosis is 1.54, skewness is 1.52 and error rate 0.076. These parameters are higher than recently developed models whose algorithms and methods accuracy is more than 90%.

Optimizing hybrid neural networks for precise COVID-19 mRNA vaccine degradation prediction

Conventional hybrid models often miss an essential factor that can lead to less effective performance: intrinsic sequence dependence when combining various neural network (NN) architectures. This study addresses this issue by highlighting the importance of sequence hybridization in NN architecture integration, aiming to improve model effectiveness. It combines NN layers—dense, long short-term memory (LSTM), and gated recurrent unit (GRU)—using the Keras Sequential API for defining the architecture. To provide better context, bidirectional LSTM (BiLSTM) and bidirectional GRU (BiGRU) replace their unidirectional counterparts, enhancing the models through bidirectional structures. Out of 25 NN models tested, 18 four-layer hybrid NN models consist of one-quarter dense layer and the rest BiLSTM and BiGRU layers. These hybrid NN models undergo supervised learning regression analysis, with mean column-wise root mean square error (MCRMSE) as the performance metric. The results show that each hybrid NN model produces unique outcomes based on its specific hybrid sequence. The Hybrid_LGSS model performs better than existing three-layer BiLSTM networks in predictive accuracy and shows lower overfitting (MCRMSEs of 0.0749 and 0.0767 for training and validation, respectively). This indicates that the optimal hybridization sequence is crucial for achieving a balance between performance and simplicity. In summary, this research could help vaccinologists develop better mRNA vaccines and provide data analysts with new insights for improvement.

Artificial Intelligence Applications in Neonatal Critical Care: A Scoping Review

Background: The development of artificial intelligence (AI) approaches impacted drug discovery, medical imaging, customized diagnostics, and therapeutics. Medicine will be transformed by AI. One such area of medicine where AI is significantly improving care is neonatology. Objective: The objective of this scoping review is to explore the applications of AI in neonatal critical care and its outcome. Methods: Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines, a scoping review was conducted utilizing the Web of Science, MEDLINE (PubMed), and Scopus databases. The search was limited to full-text publications on AI applications in neonatal critical care that were published between January 1, 2019, and December 31, 2023. Articles specifically addressing the application of AI in neonatal care have been considered within the scope of this review. At least three reviewers had independently executed the screening, data abstraction, and exploration. Results: Database searches yielded 631 articles, of which 11 met the inclusion criteria. The research encompassed extensive AI applications in neonatal critical care, employed for prognosis, diagnosis, and therapy strategizing. Artificial neural networks, machine learning, deep learning, and shallow hybrid neural networks were the commonly utilized AI techniques (neonatal critical care). These methods were applied to screen for inborn metabolic abnormalities, predict various outcomes, including death and sepsis, identify diseases such as sepsis, and assess neurodevelopmental outcomes in preterm newborns, helping plan several medical treatments. The included research demonstrated encouraging outcomes when using AI in neonatal critical care. Conclusions: AI-driven electronic arrangements upgrade neonatal basic care by improving risk forecast, promising critical commitments to future health care. Be that as it may, careful appraisal, evidence-based considers, and determination of safety, ethics, and information straightforwardness issues are essential before implementation. Acceptance by administrative bodies and the therapeutic community pivots on tending to these concerns.

Energy price prediction based on decomposed price dynamics: A parallel neural network approach

The growing uncertainties of the world due to geographic tensions, and weather conditions challenge the traditional method to achieve a reliable price prediction of energy future for both asset pricing and risk management. Following the initial success of deep learning models in energy price prediction, we attempt to establish a better architecture of neural networks to improve the prediction accuracy. We propose a novel Parallel Hybrid Neural Network (PHNN) model that utilizes independent sub-networks to effectively capture the distinct features of various sequences. Empirical results demonstrate that the PHNN model exhibits a significant performance enhancement of 16.68%, 14.09%, and 2.34% over the EMD–LSTM, the informer model, and the single LSTM model, respectively. In particular, the PHNN outperforms the single LSTM, which is trained on the same inputs, by 2.34% overall while by a remarkable 4.11% during event periods. This suggests that the PHNN derives notable benefits from its distinct architecture, particularly during the initial phase of extreme events characterized by significant price trend changes. Additionally, the study explores the potential benefits of incorporating additional event-tracking indicators for energy futures price forecasting. However, the findings suggest that these indicators may not consistently provide effective complementary information.

DEEP AND MACHINE LEARNING MODELS FOR RECOGNIZING STATIC AND DYNAMIC GESTURES OF THE KAZAKH ALPHABET

Currently, an increasing amount of research is directed towards solving tasks using computer vision libraries and artificial intelligence tools. Most common are the solutions and approaches utilizing machine and deep learning models of artificial neural networks for recognizing gestures of the Kazakh sign language based on supervised learning methods and deep learning for processing sequential data. The research object is the Kazakh sign language alphabet aimed at facilitating communication for individuals with limited abilities. The research subject comprises machine learning methods and models of artificial neural networks and deep learning for gesture classification and recognition. The research areas encompass Machine Learning, Deep Learning, Neural Networks, and Computer Vision. The main challenge lies in recognizing dynamic hand gestures. In the Kazakh sign language alphabet, there are 42 letters, with 12 of them being dynamic. Processing, capturing, and recognizing gestures in motion, particularly in dynamics, pose a highly complex task. It is imperative to employ modern technologies and unconventional approaches by combining various recognition methods/algorithms to develop and construct a hybrid neural network model for gesture recognition. Gesture recognition is a classification task, which is one of the directions of pattern recognition. The fundamental basis of recognition is the theory of pattern recognition. The paper discusses pattern recognition systems, the environment and application areas of these systems, and the requirements for their development and improvement. It presents tasks such as license plate recognition, facial recognition, and gesture recognition. The field of computer vision in image recognition, specifically hand gestures, is also addressed. The development of software will enable the testing of the trained model's effectiveness and its application for laboratory purposes, allowing for adjustments to improve the model.

The efficiency assessment of using hybrid neural networks for the detection of forged audio data in socially oriented systems

The subject of the research is the problem of detecting falsified data, in particular in audio format, in socially oriented systems. The goal of the work is to develop an effective model based on recurrent and convolutional neural networks for determining the fact of forgery of sound data, using MapReduce technology for parallelization. The article addresses the following tasks: determining the features of audio in socially-oriented systems, conducting an analysis of algorithms for processing audio information both in the form of text and in the form of a signal, forming a list of target architectures of neural networks and revealing the features of their implementation, conducting an experimental test of effectiveness selected approaches. The following methods used are – analytical and inductive method for determining the target set of neural network architectures; expert assessment for the formation of the most influential efficiency factors; experimental, multi-criteria evaluation and statistical methods of data augmentation to determine the most effective model. The following results were obtained: an audio data reprocessing algorithm was developed for the possibility of using recurrent and convolutional networks. Several approaches to data classification using augmentation based on vector autoregression and MapReduce parallelization technology have been implemented. It was determined that the most effective model for the multi-criteria selection problem is a combination of a bidirectional recurrent neural network with support for short- and long-term memory with several convolutional networks. The advantages of using MapReduce technology to optimize training time and data processing are shown, and a set of open questions for further research and applied implementation is defined. Conclusions: he application of an analytical and inductive approach followed by experimental verification made it possible to develop an effective (with an accuracy of more than 96%) a mechanism for detecting fabricated data both in the form of a signal and in text form. The obtained result makes it possible to assert the feasibility of implementing the proposed approach, and, accordingly, makes it possible to reduce the influence of such information in socially oriented systems, especially during crisis events.

AI-Personalized Elegance: Optimizing Your Appearance with Smart Hairstyle Recommendations

Many machine learning algorithms have been introduced to solve different types of problems. Recently, many of these algorithms have been applied to deep architecture models and showed very impressive performances. In general, deep architecture models suffer from the over-fitting problem when there is a small number of training data. In this article the attempt is made to remedy this problem in deep architecture with regularization techniques including overlap pooling and flipped image augmentation and dropout; the authors also compared a deep structure model (convolutional neural network (CNN)) with shallow structure models (support vector machine and artificial neural network with one hidden layer) on a small dataset. It was statistically confirmed that the shallow models achieved better performance than the deep model that did not use a regularization technique. Faces represent complex multidimensional meaningful visual stimuli and developing a computational model for face recognition is difficult. The authors present a hybrid neural-network solution which compares favorably with other methods. Keyword Hair style recommendation; Face recognition; Machine learning; Software Development

Hybrid neural network wind speed prediction based on two-level decomposition and weighted averaging

Hybrid neural network wind speed prediction based on two-level decomposition and weighted averaging

A pyramid Gaussian pooling based CNN and transformer hybrid network for smoke segmentation

AbstractVisual smoke semantic segmentation is a challenging task due to semi‐transparency, variable shapes, and complex textures of smoke. To improve segmentation performance, a convolutional neural network and transformer hybrid network are proposed based on pyramid Gaussian pooling (PGP) for smoke segmentation. In order to utilize low‐pass filtering to suppress noise, a PGP method is designed. Then, the output of PGP is reshaped to construct a set of visual tokens for transformers, thus a PGP‐transformer module is presented to make full use of the self‐attention mechanism. Finally, the PGP‐transformer module is inserted into the U‐shaped architecture with skip connections. A large number of experiments have proved that the method is significantly superior to existing state‐of‐the‐art algorithms on virtual and real smoke datasets, and ablation experiments have also verified the effectiveness of the proposed modules.

Named Entity Recognition In Electronic Medical Records Based On Hybrid Neural Network And Transformer

The development of artificial intelligence in the field of health encourages the use of electronic medical records in all health facilities to record health services provided to patients. For hospitals, extracting information from electronic medical records can make it easier for management to make clinical decisions and for researchers to obtain data for research in the medical and nursing fields. The research builds a model of named entity recognition in electronic medical records based on hybrid neural networks, bidirectional encoder representations from transformers, and setting hyperparameters to get the highest accuracy. The research data set is processed from an initial examination form of an adult patient in a hospital onto an electronic medical record in 2022, and the data is pre-processed. Next, perform the entity-tagging phase on the text and divide 70% of the training datasets by 30% of the testing datasets. Training and evaluation of models built using the confusion matrix method. The results of this study show that the entity identification model called bidirectional encoder representations from transformers consistently outperforms the neural network-based entity recognition model in any evaluation metric. The abbreviation of the bi-directional encoder representation of transformer has very high precision, recall, and f1-score values, demonstrating its ability to recognise entities very well. In this study, although the model named Entity Recognition based on neural networks also has high accuracy, the low precision and recall values indicate that this model may have difficulty recognising entities accurately.