Deep Belief Network Research Articles

Leveraging Marine Predators Algorithm with Deep Learning for Lung and Colon Cancer Diagnosis.

Cancer is a deadly disease caused by various biochemical abnormalities and genetic diseases. Colon and lung cancer have developed as two major causes of disability and death in human beings. The histopathological detection of these malignancies is a vital element in determining the optimal solution. Timely and initial diagnosis of the sickness on either front diminishes the possibility of death. Deep learning (DL) and machine learning (ML) methods are used to hasten such cancer recognition, allowing the research community to examine more patients in a much shorter period and at a less cost. This study introduces a marine predator's algorithm with deep learning as a lung and colon cancer classification (MPADL-LC3) technique. The presented MPADL-LC3 technique aims to properly discriminate different types of lung and colon cancer on histopathological images. To accomplish this, the MPADL-LC3 technique employs CLAHE-based contrast enhancement as a pre-processing step. In addition, the MPADL-LC3 technique applies MobileNet to derive feature vector generation. Meanwhile, the MPADL-LC3 technique employs MPA as a hyperparameter optimizer. Furthermore, deep belief networks (DBN) can be applied for lung and color classification. The simulation values of the MPADL-LC3 technique were examined on benchmark datasets. The comparison study highlighted the enhanced outcomes of the MPADL-LC3 system in terms of different measures.

Structural Design for Roll-Formed Aluminium Alloy Perforated Channels Subjected to Interior-Two-Flange Web Crippling: Experimental Tests, Numerical Simulation, and Neural Network

This study analyses the interior-two-flange (ITF) web crippling strength of roll-formed aluminum alloy lipped channels (RA channels) with web holes employing experimental testing, numerical modeling, and deep neural network (i.e., Deep belief Network, DBN). A total of 30 experimental tests on web crippling behavior were carried out, with the results utilized to validate a finite element (FE) model, developed in this study. The experimental results were compared to the data produced by the validated FE model, which was then used to train the DBN model. The results of the DBN prediction were shown to be around 5% more conservative than the FE results. In order to evaluate the effects of associated factors on the ITF web crippling strength of RA channels, a comprehensive parametric study was conducted using the DBN. The design guidelines that are currently available in the American Iron and Steel Institute (AISI 2016), the Australian and New Zealand Standards (AS/NZS 1997; AS/NZS (2018)), and the Eurocode (CEN 2007) were found to be unreliable while determining the ITF web crippling strength of RA channels. The DBN's predictions developed new formulae for calculating the web crippling strength reduction factors. After conducting a reliability study, it was found that the developed strength reduction factor equations are reliable when calculating the ITF web crippling strength of such perforated roll-formed aluminium alloy channels.

Research on Fault Diagnosis of Hoist Dangerous Load Based on DBN

At present, hoisting accidents caused by dangerous loads such as overloading, incomplete unloading, conveyance jamming, and rope loosening still occur. Therefore, this paper proposes a fault diagnosis method for dangerous loads of hoists based on the Deep Belief Network (DBN). With the stator current of the hoisting motor as the signal source, the 31-dimensional eigenvalues of the stator current of the hoisting motor in the time, frequency, and time-frequency domains are taken as the input parameters of the deep belief network. The restricted Boltzmann machine is used for unsupervised pre-training through layer-by-layer transmission and layer-by-layer training. The BP neural network, as the top-level classifier, is used for supervised fine-tuning training to finally complete the whole training process of the depth belief network. Relying on the advantages of the depth belief network, which has fast training speed and strong feature extraction ability. It makes up for the deficiency of the traditional algorithm in feature extraction and classification separation, resulting in the loss of details, and finally realizes the overload Classification of four types of dangerous loads, i.e., unloaded, blocked, and slackened. The results show that the method based on DBN has higher identification and better diagnosis effect.

Empirical Study of Protein Feature Representation on Deep Belief Networks Trained With Small Data for Secondary Structure Prediction.

Protein secondary structure (SS) prediction is a classic problem of computational biology and is widely used in structural characterization and to infer homology. While most SS predictors have been trained on thousands of sequences, a previous approach had developed a compact model of training proteins that used a C-Alpha, C-Beta Side Chain (CABS)-algorithm derived energy based feature representation. Here, the previous approach is extended to Deep Belief Networks (DBN). Deep learning methods are notorious for requiring large datasets and there is a wide consensus that training deep models from scratch on small datasets, works poorly. By contrast, we demonstrate a simple DBN architecture containing a single hidden layer, trained only on the CB513 dataset. Testing on an independent set of G Switch proteins improved the Q 3 score of the previous compact model by almost 3%. The findings are further confirmed by comparison to several deep learning models which are trained on thousands of proteins. Finally, the DBN performance is also compared with Position Specific Scoring Matrix (PSSM)-profile based feature representation. The importance of (i) structural information in protein feature representation and (ii) complementary small dataset learning approaches for detection of structural fold switching are demonstrated.

A design of movie script generation based on natural language processing by optimized ensemble deep learning with heuristic algorithm

Movies offer users a huge range of visual information like attractive stories. The traditional approaches have demonstrated that knowing about movie stories via only visual information is complicated. The natural data of graphical texts play a significant task in portraying information in various domains like entertainment, education, communication, etc. Similarly, script generation by considering the earlier dialogue is more complex owing to the inherent nature of the text. Text identification in natural data considers script identification that needs localization of text. Especially for natural scene data, script generation is a complicated task due to different background/foreground components. The texts comprise varied textures, variations in size, orientation, colors, and fonts. It is essential to have a system that produces scripts or stories from a storyline automatically. However, this research is not emerging nowadays. While considering the dialogue systems would also help drive dialogues through a dialogue plan. Hence, a new movie script generation model is suggested through processing the movie text data. Initially, the text data is collected related to a different movie that consists of characters, scenes, genre, location, etc. Secondly, the data pre-processing is carried out to enhance data quality. Further, the significant features are extracted from the pre-processed data through Term Frequency-Inverse Document Frequency (TFIDF) and word2vector. The deep features are extracted to get the noteworthy features using a Deep Belief Network (DBN) from RBN layers. Finally, the deep features are given to the Ensemble-based Movie Scrip Generation (EMCG), where the Optimized hybrid script generation process using ensemble learning is performed by Bidirectional Long Short-Term Memory (Bi-LSTM), Generative Pre-Trained Transformer version 3 (GPT3), and GPT Neo X models, where the parameters of deep learning algorithms are optimized using the Adaptively Improved Cat and Mouse-based Optimizer (AI-CMO) algorithm. Here, the outcomes are attained through taking averaging among the classified outcomes. The standard performance measures are used for evaluating the effectiveness of the proposed method.

A flexible RUL prediction method based on poly-cell LSTM with applications to lithium battery data

Remaining useful life estimation is one of the major prognostic targets in the battery management system. Since the degradation process of the lithium battery is a complex nonlinear process, including temporary capacity regeneration as well as noise interference, the deep learning model is a popular method to discuss the remaining useful life prediction. However, most deep learning networks, including the convolutional neural network, deep belief network, recurrent neural network, and their variants, cannot process input data with different update states according to data importance. In order to improve the prediction accuracy of RUL, this paper proposes a novel model named Poly-Cell Long Short-Term Memory Network, which adds a hierarchical division unit and a poly-cell unit The model determines the importance of the input data through the hierarchical division unit and then uses the poly-cell unit to update the cell state according to the features’ importance.The proposed RUL prediction method is illustrated by using the lithium battery data. The experimental results show the efficiency of the proposed prediction approach. • The proposed PCLSTM uses degradation information more effectively. • The PCLSTM eases the computational burden. • The proposed PCLSTM is equipped with flexibility and robustness. • The effectiveness of the proposed method is illustrated by analyzing the lithium battery data.

WASSERSTEIN GAN-GRADIENT PENALTY WITH DEEP TRANSFER LEARNING FOR 3D MRI ALZHEIMER DISEASE CLASSIFICATION

MRI scans for Alzheimer's disease (AD) detection are popular. Recent computer vision (CV) and deep learning (DL) models help construct effective computer assisted diagnosis (CAD) models for AD detection and categorization. Due to their dependence on huge training datasets and effective hyper parameter tuning procedures, most models failed. Transfer learning adjusts the final fully linked layers to use trained DL models on smaller datasets. It handles picture categorization problems well. This research introduces a Wasserstein GAN-Gradient Penalty with Deep Transfer Learning (WGANGP-DTL)-based AD classification model on 3D MRI data. WGANGP-DTL aims to accurately identify and classify AD. WGANGP increases dataset size. The WGANGP-DTL model pre-processes MRI images using image enhancement and segments them using 3DSFCM. Feature extraction uses ant lion optimizer (ALO) with Inception v3 model. For classifying AD, deep belief network (DBN) model is used. The WGANGP-DTL model is experimentally validated using benchmark 3D MRI datasets. WGANGP-DTL outperformed recent methods in experiments.

Personalization of a tourism recommender system based on users similarity and the use of deep belief network

Personalization of a tourism recommender system based on users similarity and the use of deep belief network

Automated climate prediction using pelican optimization based hybrid deep belief network for Smart Agriculture

A significant part of the economies of several nations throughout the world, including India, where agriculture accounts for 16% of the total economy. Due to their dynamic and turbulent natures, climate prediction in this region is the most difficult because statistical methodologies can't provide accurate predictions. Typically, weather forecasts were made using extraordinarily complex physics techniques that took into account various atmospheric conditions over a long time. These circumstances were usually unstable as a result of weather system disturbances, which led to the development of techniques for making unreliable predictions. In this study, an Automated Climate Prediction for Smart Agriculture is developed utilizing a Pelican Optimization-based Hybrid Deep Belief Network (ACP-POHDBN). The purpose of the ACP-POHDBN approach is to identify the appropriate meteorological conditions. The acp-POHDBN method uses the test data to calculate the appropriate weather conditions. Pre-processing, prediction, and hyperparameter tuning are the three key steps that make up the proposed ACP-POHDBN approach. The min-max normalization procedure is the primary procedure used by the ACP-POHDBN model to convert the meteorological data into a standard format. In addition to pre-processing data, the Deep Belief Network (DBN) model is used to forecast weather conditions. Finally, the DBN method's hyperparameters (epoch, batch size, and learning rate) may be properly tuned using the POA-based hyperparameter tuning technique. The development of a Pelican Optimization Algorithm (POA)-based hyperparameter optimizer for the process of predicting the climate demonstrates the originality of the study. The proposed algorithm over other recent approaches with maximum accuracy of 95.03%, sensitivity of 95.03%, specificity of 95.03%, and F-score of 95.03%.

Convolutional Neural Network-Based Machine Vision for Non-Destructive Detection of Flooding in Packed Columns.

In chemical processes, packed columns are frequently employed in various unit operations. However, the flow rates of gas and liquid in these columns are often constrained by the risk of flooding. To ensure the safe and efficient operation of packed columns, it is crucial to detect flooding in real time. Conventional flooding monitoring methods rely heavily on manual visual inspections or indirect information from process variables, which limit the real-time accuracy of results. To address this challenge, we proposed a convolutional neural network (CNN)-based machine vision approach for non-destructive detection of flooding in packed columns. Real-time images of the packed column were captured using a digital camera and analyzed with a CNN model, which was been trained on a dataset of recorded images to identify flooding. The proposed approach was compared with deep belief networks and an integrated approach of principal component analysis and support vector machines. The feasibility and advantages of the proposed method were demonstrated through experiments on a real packed column. The results showed that the proposed method provides a real-time pre-alarm approach for detecting flooding, enabling process engineers to quickly respond to potential flooding events.

Compared between support vector machine (SVM) and deep belief network (DBN) for multi-classification of Raman spectroscopy for cervical diseases

In this study, a minimally invasive test method for cervical cancer in vitro was proposed by comparing Raman spectroscopy with support vector machine (SVM) model and deep belief network (DBN) model. The serum Raman spectra of cervical cancer, hysteromyoma, and healthy people were collected. After data processing, SVM classification model and DBN classification model were built respectively. The experimental results show that when the DBN network algorithm is used, the sample test set can be divided accurately and the result of cross-validation is ideal. Compared with the traditional SVM algorithm, this method firstly screened the effective feature matrix from the data, and then classified the data. With high efficiency and accuracy, based on 445 samples collected, this method improved the accuracy by 13.93%±2.47% compared with the SVM method, and provided a new direction and idea for the in vitro diagnosis of cervical diseases.

Deep learning-based EEG emotion recognition: Current trends and future perspectives.

Automatic electroencephalogram (EEG) emotion recognition is a challenging component of human-computer interaction (HCI). Inspired by the powerful feature learning ability of recently-emerged deep learning techniques, various advanced deep learning models have been employed increasingly to learn high-level feature representations for EEG emotion recognition. This paper aims to provide an up-to-date and comprehensive survey of EEG emotion recognition, especially for various deep learning techniques in this area. We provide the preliminaries and basic knowledge in the literature. We review EEG emotion recognition benchmark data sets briefly. We review deep learning techniques in details, including deep belief networks, convolutional neural networks, and recurrent neural networks. We describe the state-of-the-art applications of deep learning techniques for EEG emotion recognition in detail. We analyze the challenges and opportunities in this field and point out its future directions.

Mass estimation of loose parts based on virtual experiment

Mass estimation of loose parts in primary coolant system is an important task of loose parts monitoring system. However, it is difficult to carry out drop tests of loose parts in primary coolant system so as to provide the field data for developing a mass estimation model. A virtual experiment approach based on finite element simulation is proposed, which contains three steps: (1) finite element modeling and updating of impacted structural component; (2) simulating the drop tests of loose parts on structural component to generate the virtual experimental data; (3) establishing a mass estimation model using deep belief network based on the virtual experimental data. The flat plate steel ball drop tests are performed to demonstrate the presented approach. The results show that the prediction accuracy of mass estimation model built based on virtual experimental data is quite close to that of the model built based on drop test data.

Phonocardiogram signal classification for the detection of heart valve diseases using robust conglomerated models

The diagnosis of cardiovascular diseases is quite important in the field of medical community. An important physiological signal of human body is heart sound and it arises due to the blood turbulence and pulsing of cardiac structures. For the early diagnosis of heart diseases, the analysis of heart sounds play an important role as they contain a huge quantity of pathological information associated with heart. To detect heart sounds, Phonocardiogram (PCG) is used as it is a highly useful and non-invasive technique and can be easily analyzed well. In this paper, some efficient models are proposed for the classification of PCG signals. Two important and robust conglomerated models are proposed initially, wherein the first strategy utilizes the concept of semi-supervised Non-negative Matrix Factorization (NMF) along with Brain Storming (BS) optimization algorithm and an advanced version of BS termed as Advanced BS (ABS) is proposed and then it is merged with Genetic Programming (GP) so that new algorithms such as BS-GP and ABS-GP are formed and finally the features selected through it are fed to classification through machine learning. The second strategy utilizes the concept of using three dimensionality reduction techniques along with Fuzzy C-means (FCM) clustering and then an Advanced Sine-Cosine (ASC) optimization algorithm with three different modifications is proposed for the purpose of feature selection and finally it is classified. Deep learning techniques were also employed in the study such as the usage of an Attention based Bidirectional Long Short-Term Memory (A-BLSTM), Ordinal Variational Autoencoder (O-VAE), Conditional Variational Autoencoders (CVAE), Hyperspherical CVAE (H-CVAE) and the Restricted Boltzmann Machine based Deep Belief Network (RBM-DBN) for the classification of PCG signals. The experiment is conducted on a publicly available dataset and results show that a high classification accuracy of 95.39% is obtained for the semi-supervised NMF concept with ABS-GP technique and Support Vector Machine (SVM) classifier.

A skin cancer diagnosis system for dermoscopy images according to deep training and metaheuristics

One of the most common kinds of cancer in the United States is skin cancer. This kind of cancer occurs when the cells that make up the skin layer grow uncontrollably and divide rapidly. To avoid the skin cancer harms, it is better to detect it in the early stages. Due to complexity of the diagnosis, some times, the experts provide an incorrect diagnosis about this cancer. Therefore, proposing a computer-aided system, not as an independent system, but as an auxiliary tool along with the physicians, can help them in more accurate detection of this kind of cancer. This study introduced an optimal and efficient computer-assisted system for skin cancer analysis. Here, deep training and metaheuristic techniques are used for this purpose. The main idea is to provide a Deep Belief Network (DBN), based on an improved metaheuristic technique, called Modified Electromagnetic Field Optimization Algorithm (MEFOA) to provide a strong diagnosis system for the images of skin cancer. The function of the proposed method is proved through a comparison between its results and some other related works.

Enhancement in web accessibility for visually impaired people using hybrid deep belief network –bald eagle search

Enhancement in web accessibility for visually impaired people using hybrid deep belief network –bald eagle search

Microblog Emotion Analysis Using Improved DBN Under Spark Platform

In order to solve the problems that traditional single-machine methods find it difficult to complete the task of emotion classification quickly, and the time efficiency and scalability are not high; a microblog emotion analysis method using improved deep belief network (DBN) under Spark platform is proposed. First, the Hadoop distributed file system is used to realize the distributed storage of text data, and the preprocessed data and emotion dictionary are converted into word vector representation based on the continuous bag-of-words model. Then, an improved DBN model is constructed by combining the adaptive learning method of DBN with the active learning method, and it is applied to the learning analysis of text word vectors. Finally, the data parallel optimization of the improved DBN model is realized, based on Spark platform to accurately and quickly obtain the emotion types of microblog texts. The experimental analysis of the proposed method based on the microblog text data set shows that the classification accuracy is more than 94%.

Sparse non-convex regularization based explainable DBN in the analysis of brain abnormalities in schizophrenia

Deep belief networks have been widely used in medical image analysis. However, the high-dimensional but small-sample-size characteristic of medical image data makes the model prone to dimensional disaster and overfitting. Meanwhile, the traditional DBN is driven by performance and ignores the explainability which is important for medical image analysis. In this paper, a sparse non-convex based explainable deep belief network is proposed by combining DBN with non-convex sparsity learning. For sparsity, the non-convex regularization and Kullback–Leibler divergence penalty are embedded into DBN to obtain the sparse connection and sparse response representation of the network. It effectively reduces the complexity of the model and improves the generalization ability of the model. Considering explainability, the crucial features for decision-making are selected through the feature back-selection based on the row norm of each layer’s weight after network training. We apply the model to schizophrenia data and demonstrate it achieves the best performance among several typical feature selection models. It reveals 28 functional connections highly correlated with schizophrenia, which provides an effective foundation for the treatment and prevention of schizophrenia and methodological assurance for similar brain disorders.

Application of deep learning in recognition of accrued earnings management

We choose the sample data in Chinese capital market to compare the measurement effect of earnings management with Deep Belief Network, Deep Convolution Generative Adversarial Network, Generalized Regression Neural Network and modified Jones model by performance. We find that Deep Belief Network has the best effect, while Deep Convolution Generative Adversarial Network has no significant advantage, and the measurement effect of Generalized Regression Neural Network and modified Jones model have little difference. This paper provides empirical evidence that neural networks based on deep learning technology and other artificial intelligence technologies can be widely applied to measure earnings management in the future.

An integrated 3D-sparse deep belief network with enriched seagull optimization algorithm for liver segmentation

An integrated 3D-sparse deep belief network with enriched seagull optimization algorithm for liver segmentation