Deep Belief Network

Traditional, threshold-based "pass/fail" maintenance philosophies for power transformers are fundamentally reactive, often failing to identify assets on a trajectory toward failure and thereby creating significant operational and financial risk. This paper analyzes two distinct, real-world case studies to demonstrate the strategic superiority of a predictive asset management paradigm founded on data trend analysis and the targeted use of advanced diagnostics. The first case study examines an aging 30-year-old transformer (Legacy PTR) whose end-of-life condition, first suggested by accelerating negative trends in routine electrical and oil tests, was definitively confirmed by Furan analysis (3 ppm 2-FAL). This diagnostic confirmation allowed for a planned decommissioning, averting a probable in-service catastrophic failure that superficial, symptom-based interventions would have missed. The second case study investigates a recently repaired transformer (Reborn PTR) that presented a deceptively healthy profile on standard metrics like Breakdown Voltage (BDV) following an oil filtration process. However, a more sensitive Interfacial Tension (IFT) analysis revealed severe, unaddressed chemical contamination, highlighting the profound danger of relying on an incomplete diagnostic picture. Synthesizing the lessons from these cases, this paper argues that a robust, modern asset management strategy must prioritize the analysis of degradation velocity (trends) over static absolute values and must strategically deploy advanced diagnostics to uncover rootcause conditions. This integrated approach enables proactive, risk-informed decisions that enhance grid reliability, improve safety, and optimize the total lifecycle cost of critical assets.

While enabling remote management and efficiency improvements, the infrastructure of the smart city becomes able to advance due to the consequences of the internet of things (IoT). The development of IoT in the fields of agriculture, robotics, transportation, computerization, and manufacturing. Based on the serious infrastructure environments, smart revolutions and digital transformation play an important role. According to various perspectives on issues of privacy and security, the challenge is heterogeneous data handling from various devices of IoT. The critical IoT infrastructure with its regular operations is jeopardized by the sensor communication among both IoT devices depending upon the attacker targets. This research suggested a novel deep belief network (DBN) and a secured data dissemination structure based on blockchain to address the issues of privacy and security infrastructures. The non-local means filter performs pre-processing and the feature selection is achieved using the improved crystal structure (ICS) algorithm. The DBN model for the classification of attack and non-attack data. For the non-attacked data, the security is offered via a blockchain network incorporated with the interplanetary file system.

Lung cancer is an early lung cancer. The world's greatest cause of death is lung cancer. Lung cancer typically does not cause signs or symptoms in its early stages. In order to reduce the mortality rate early detection of lung cancer is required. Computed Tomography (CT) lung images classification for nodule detection is discussed in this study. At first energy features are used for feature extraction. Then maximum likelihood classifier is used for classification. The experimental results show the performance of pulmonary nodules classification in terms of accuracy.

Plasmodium is a type of unicellular eukaryote compulsory for vertebrates or insect parasites. The early diagnosis is required for malaria. In this study, the automatic classification of malaria system is discussed. Initially, the input images are given to gaussian filter, then energy feature is used for feature extraction and K-Nearest Neighbor (KNN) classifier is used for classification. The performance of malaria system produces the specificity of classification 93%using KNN classifier.

Malaria is an infectious disease transmitted by mosquitoes that affects humans and other animals. Malaria is responsible for the effects of fever, tiredness, vomiting and headaches. Yellow skin, convulsions, a coma, or death can lead to severe cases. Symptoms usually start 10-15 days after a mosquito is bitten. The early diagnosis is required for malaria. In this study, the automatic classification of malaria system is discussed. Initially, the input images are given to Fourier–Mellin transform for feature extraction and Support Vector Machine (SVM) classifier is used for classification. The performance of malaria system produces the classification accuracy of 92%using SVM classifier.

Detecting motor imagery from electrocardiographic (ECG) signals is complex but crucial in developing advanced neuroprosthetic devices and brain-computer interface (BCI) systems. In most cases, linear models applied using conventional methods are not appropriate for the time-varying and non-linear nature represented by the ECG characteristics, resulting in weak performances. This research addresses this problem, combining Wavelet Packet Decomposition and Multi-Scale Convolutional Neural Networks to improve the feature extraction mechanism and classification accuracy. ECG data is pre-processed from the PhysioNet EEG Motor Movement/Imagery Dataset to remove noise and standardize signals. WPD is thus applied to decompose the signals into detailed frequency components to be input as features in the proposed Multi-Scale CNN. Different kernel sizes are implemented in these parallel convolutional layers to learn complicated features at various hierarchical resolutions. The proposed architecture is evaluated using performance parameters such as accuracy 92%, precision 89%, recall 93%, F1 score 91%, and ROC-AUC 95%. These results showed that the model outperformed the earlier-used traditional methods, such as Support Vector Machines (SVM) and Random Forests, better-detecting motor imagery. This research emphasizes the integrative power of advanced signal processing techniques with deep learning in analyzing biomedical signals, providing a powerful solution to advancing neuroprosthetic and BCI technologies. Povzetek: Študija dokazuje učinkovitost kombinacije obdelave signalov in globokega učenja za analizo biomedicinskih signalov. Uporabljena je valčna paketna dekompozicije in večskalna konvolucijska nevronska mreža za detekcijo motorične imaginacije v signalih EKG.

Millions of people depend on coconut palms for their food and livelihoods, making them one of the most essential crops in tropical countries. However, Diseases may significantly reduce the output of coconut trees and possibly result in their death. To overcome this, a novel golden jackal optimized disease detection in COCOnut tree (GOD-COCO) has been proposed for detecting diseases in coconut trees. First, the input dataset images are preprocessed in pre-processing image rotation, image rescaling, and image resizing, and the enhanced images are gathered. The enhanced images are segmented using the PSP-Net. From the segmented images, the features are extracted using the Dense-Net. Then the features needed are selected using the golden jackal optimization algorithm (GJOA). Finally, the deep belief network (DBN) classifier classifies whether it is normal or abnormal. The experimental analysis of the proposed GOD-COC has been evaluated using the Plant Pathology datasets based on the accuracy, precision, and recall standards. By this, the proposed GOD-COCO achieves an accuracy rate of 99.31% and it achieves an overall accuracy rate of 0.77%, 0.31% and 1.17% by the existing methods such as AIE-CTDDC, DL-WDM, and CLS. Similarly, the proposed GOD-COCO model takes less time, 1.13 milliseconds to detect the disease, than the existing methods, which take 3.04, 2.5, and 2.67 milliseconds, respectively.

Detecting buildings from satellite imagery presents challenges related to computational efficiency, model adaptation, and occlusion. This paper introduces a novel method called the Secant Deep Belief Network-Hyperbolic Cosine Whale Optimization (SDBN-HCWO) for building detection in satellite images. The research utilizes the SDBN-HCWO method to enhance building detection accuracy in satellite images. It addresses challenges like computational efficiency, occlusion, and dataset adaptation. The method integrates a multi-layer structure, including Hyperbolic Cosine Prey Encircling for edge identification, Shrinking Encircle for optimal edge linking, and Secant Object Detection for accurate identification. Additionally, a Densely Connected Convolutional Network (DCCN) and Depth-wise Separable Convolution (DSC) optimize feature extraction, reducing computational costs. The model is evaluated on both quantitative and qualitative metrics, ensuring high accuracy and low false positive rates. The research findings demonstrate that the SDBN-HCWO method significantly improves building detection accuracy in satellite imagery. It enhances detection efficiency by integrating Discrete Latent Deep Reinforcement Learning and a bubble-net mechanism, reducing false positives by 58 %. The model outperforms conventional approaches, achieving an 18 % increase in PSNR, 34 % rise in CA, and 19 % reduction in training time. High AP scores (90.40 %-92.67 %) confirm its reliability, though challenges persist in medium-damage detection. It surpasses YOLOv3, YOLOv4, and Faster R-CNN in accuracy and efficiency. This research significantly advances building detection in satellite imagery, facilitating more accurate urban planning, disaster response, and environmental monitoring.

In this study, attention-based fusions of CNN and RNN indeed reached an interesting performance with identifying the fetal brain MRIs. Various Deep Learning and Transfer learning approaches were proposed for improved accuracy and detecting abnormal fetal neurological development. Fetal brain MRI acquisition and processing fetches fetal brain MRI from the fetal brain MRI database randomly, it resizes them to 224 × 224. This TensorFlow function normalizes and pre-processes the images and ensures that they are compatible with ResNet50 architecture. ResNet50 CNN architecture that is pre-trained with ImageNet weights is removing the fully connected layers because it extracts high-dimensional spatial features from the MRI images. These resulting feature maps will then be reshaped into a sequence of RNN input feature vectors. The CNN features extracted were passed to Bidirectional Long Short-Term Memory (LSTM) network and Gated recurrent units (GRU) for sequence-level dependency mapping. In AEC, an attention mechanism pays attention to relevant time steps which are utilized for classification. Fully connected layers, with dropout to prevent overfitting, are mapped on the output of RNN and the output is classified using a sigmoid activation function into healthy vs. abnormal fetal brain. The CNN-RNN architecture was trained with binary cross-entropy loss functions as well as the Adam optimizer for the randomly generated labels. Then train on these for 10 epochs on batch size = 8, and see you get a decent loss and accuracy. We produce visualizations to figure out where the model is performing correctly and inaccurately,

A precise tool wear monitoring model is essential for manufacturing to ensure reliability and efficiency. This study aims to analyze and monitor the condition of small-sized cutting tools during end-milling operations based on direct and indirect approaches in machining AISI H13 alloy steel. The tool condition monitoring classification method was achieved by integrating the wavelet transform method for multiresolution analyses with a hybrid deep learning algorithm. A new approach combines maximal overlap discrete wavelet transform (MODWT) for signal preprocessing with a hybrid deep learning model that includes convolutional neural network (CNN) and bidirectional long-short term memory (BiLSTM) algorithms to improve tool wear identification accuracy and efficiency. In the present work, tool wear conditions are classified into five classes: normal tool, slight wear, moderate wear, high wear, and severe wear. The proposed model's performance was evaluated by comparing its identification accuracy to other common machine and deep learning models. This evaluation was conducted through a case study that utilized a dataset obtained from a milling test. The proposed classification model is more accurate than other machine and deep learning models. During training, it achieves a classification accuracy of 98.96%, and the overall testing accuracy is 94.07%. The effectiveness and adaptability of the proposed method in tool condition monitoring applications are noteworthy.

Machine learning is a branch of artificial intelligence. It enables computers to
automatically learn and improve from experience without being explicitly
programmed [1]. Machine learning algorithms are usually classified into
supervised, unsupervised, neural network-based, and reinforcement learning
methods. Supervised learning methods build a mathematical model from a set of
data that contains both the inputs and the desired outputs. The model is studied
to predict the output values for given inputs [1-2]. These methods can also be
used to find patterns (usually called classes) in data and assign new data points
to one of the predefined classes. The goal of unsupervised learning is to find
meaningful patterns in a set of data points that correspond closely to some
intuitive notion of similarity. Neural network-based methods belong to the class
of algorithms inspired by biological neural networks [2-4]. Neural networks
replace a simulated neuron's simple threshold function with a real-valued
differentiable function that gives the neuron the ability to output a real number.
Reinforcement learning is an area of machine learning concerned with how an
agent should take actions in an environment to maximize a reward signal.

It is of paramount importance to track the cognitive activity or cognitve attenion of the service personnel in a Prognostics and Health Monitoring (PHM) service related training or operation environment. The electroencephalography (EEG) data is one of the good candidates for cognitive activity recognition of the user. Analyzing electroencephalography (EEG) data in an unconstrained (natural) environment for understanding cognitive state and classifying human activity is a challenging task due to multiple reasons such as low signal-to-noise ratio, transient nature, lack of baseline availability and uncontrolled mixing of various tasks. This paper proposes a framework based on an emerging tool named deep learning that monitors human activity by fusing multiple EEG sensors and also selects a smaller sensor suite for a lean data collection system. Real-time classification of human activity from spatially non collocated multi-probe EEG is executed by applying deep learning techniques with...

Network representation learning and its applications have received increasing attention. Due to their various application areas, many research groups have developed a diverse range of software tools and techniques to learn representation for different types of networks. However, to the best of our knowledge, there are limited works that support representation learning for dynamic heterogeneous networks. The work presented in this demonstration paper attempts to fill the gap in this space by developing and publicly releasing an open-source Python library known as, PyDHNet, a Python Library for Dynamic Heterogeneous Network Representation Learning and Evaluation. PyDHNet consists of two main components: dynamic heterogeneous network representation learning and task-specific evaluation. In our paper, we demonstrate that PyDHNet has an extensible architecture, is easy to install (through PIP) and use, and integrates quite seamlessly with other Python libraries. We also show that the implementation for PyDHNet is efficient and enjoys a competitive execution time. • Computing methodologies → Learning latent representations; • Mathematics of computing → Graph algorithms; • Information systems → Retrieval models and ranking.

In recent years, the prevalent online services generate a sheer volume of user activity data. Service providers collect these data in order to perform client behavior analysis, and offer better and more customized services. Majority of these data can be modeled and stored as graph, such as the social graph in Facebook, user-video interaction graph in Youtube. These graphs need to evolve over time to capture the dynamics in the real world, leading to the invention of dynamic graphs. However, the temporal information embedded in the dynamic graphs brings new challenges in analyzing and deploying them. Events staleness, temporal information learning and explicit time dimension usage are some example challenges in dynamic graph learning. In order to offer a convenient reference to both the industry and academia, this survey presents the Three Stages Recurrent Temporal Learning Framework based on dynamic graph evolution theories, so as to interpret the learning of temporal information wi...

People’s attitudes, opinions, feelings and sentiments which are usually expressed in the written languages are studied by using a well known concept called the sentiment analysis. The emotions are expressed at various different levels like document, sentence and phrase level are studied by using the sentiment analysis approach. The sentiment analysis combined with the Deep learning methodologies achieves the greater classification in a larger dataset. The proposed approach and methods are Sentiment Analysis and deep belief networks, these are used to process the user reviews and to give rise to a possible classification for recommendations system for the user. The user assessment classification can be progressed by applying noise reduction or pre-processing to the system dataset. Further by the input nodes the system uses an exploration of user’s sentiments to build a feature vector. Finally, the data learning is achieved for the suggestions; by using deep belief network. The protot...

INTRODUCTION: Diabetic nephropathy is one of the complications of diabetes that causes damage to kidneys. Deep learning techniques are widely used to predict different diseases. OBJECTIVES: The main aim of this work is to develop an effective prediction model using deep learning. To get an effective model, a suitable dataset is considered that comprises of features related to diabetic nephropathy. METHODS: Deep belief network (DBN) is the proposed deep learning technique which is compared with naive bayes, CART decision tree, logistic regression and support vector machine. DBN is composed of Restricted Boltzmann Machines (RBM). The algorithms are analysed based on evaluation measures like area under PR curve, area under ROC curve, gini coefficient and jaccard index. RESULTS: After comparison of all algorithms, it was observed that DBN has performed better in terms of AUROC, gini coefficient and jaccard index with values 0.8203, 0.6406 and 0.7777 respectively. But CART obtained better value of 0.9039 only for AUPR. CONCLUSION: The proposed technique has outperformed other techniques in terms of three metrics and is identified as the best performing algorithm. Hence, it is suggested to use DBN while predicting diabetic nephropathy.

In human body, the brain is a complex organ with billions of cells that controls the overall activities of our body. The abnormal and uncontrolled multiplication of these cells leads to Brain tumour which is one of the most dangerous and aggressive second leading diseases in human beings around the world. Many Machine Learning algorithms are used with different classifiers such as Discrete Wavelet Transform, Support Vector Machine, Principal Component Analysis, etc., on Magnetic Resonance (MR) images to segment and classify tumour types. These methods are used only to predict whether the patient has brain tumour or not and also never consider the morphology and histological nature of brain tumour types. This method provides accuracy based on MR image quality, features, contrast, brightness etc. These methods show upto 95-97% of accuracy. In the next decades, Deep Learning (DL) architectures such as Deep Belief Network (DBN), Deep Neural Network (DNN), and Convolutional Neural Network (CNN) are introduced used and they are mainly focused on the fields as image recognition, natural processing, video recognition, etc. The CNN model used to recognize the tumour upto 96% of accuracy. To enhance the accuracy in brain tumour detection, a Sequential Recurrent Neural Network (SRNN) model is implemented in which instead of MR images, symptoms are considered to predict brain tumour. These symptoms are obtained through Optimal Two Phase Feature Selection (OTPFS) technique. The proposed SRNN-OTPFS model is compared with machine learning algorithms with morphology and histology fixed as a target feature. The proposed model increases the accuracy of 2 % compared to the machine learning model in both morphological and histological aspects.

Related product recommendation (RPR) is pivotal to the success of any e-commerce service. In this paper, we deal with the problem of recommending related products i.e., given a query product, we would like to suggest top-k products that have high likelihood to be bought together with it. Our problem implicitly assumes asymmetry i.e., for a phone, we would like to recommend a suitable phone case, but for a phone case, it may not be apt to recommend a phone because customers typically would purchase a phone case only while owning a phone. We also do not limit ourselves to complementary or substitute product recommendation. For example, for a specific night wear t-shirt, we can suggest similar t-shirts as well as track pants. So, the notion of relatedness is subjective to the query product and dependent on customer preferences. Further, various factors such as product price, availability lead to presence of selection bias in the historical purchase data, that needs to be controlled for while training related product recommendations model. These challenges are orthogonal to each other deeming our problem nontrivial. To address these, we propose DAEMON, a novel Graph Neural Network (GNN) based framework for related product recommendation, wherein the problem is formulated as a node recommendation task on a directed product graph. In order to capture product asymmetry, we employ an asymmetric loss function and learn dual embeddings for each product, by appropriately aggregating features from its neighborhood. DAEMON leverages multi-modal data sources such as catalog metadata, browse behavioral logs to mitigate selection bias and generate recommendations for cold-start products. Extensive offline experiments show that DAEMON outperforms state-of-the-art baselines by 30-160% in terms of HitRate and MRR for the node recommendation task. In the case of link prediction task, DAEMON presents 4-16% AUC gains over state-ofthe-art baselines. DAEMON delivers significant improvement in revenue and sales as measured through an A/B experiment.

A rapid increase in heart disease has occurred in recent years, which might be the result of unhealthy food, mental stress, genetic issues, and a sedentary lifestyle. There are many advanced automated diagnosis systems for heart disease prediction proposed in recent studies, but most of them focus only on feature preprocessing, some focus on feature selection, and some only on improving the predictive accuracy. In this study, we focus on every aspect that may have an influence on the final performance of the system, i.e., to avoid overfitting and underfitting problems or to solve network configuration issues and optimization problems. We introduce an optimally configured and improved deep belief network named OCI-DBN to solve these problems and improve the performance of the system. We used the Ruzzo-Tompa approach to remove those features that are not contributing enough to improve system performance. To find an optimal network configuration, we proposed a stacked genetic algorithm that stacks two genetic algorithms to give an optimally configured DBN. An analysis of a RBM and DBN trained is performed to give an insight how the system works. Six metrics were used to evaluate the proposed method, including accuracy, sensitivity, specificity, precision, F1 score, and Matthew's correlation coefficient. The experimental results are compared with other state-of-the-art methods, and OCI-DBN shows a better performance. The validation results assure that the proposed method can provide reliable recommendations to heart disease patients by improving the accuracy of heart disease predictions by up to 94.61%.

Motivation. In medical field, particularly the cardiology, the diagnosis systems constitute the essential domain of research. In some applications, the traditional methods of classification present some limitations. The neuronal technique is considered as one of the promising algorithms to resolve such problem. . In this paper, two approaches of the Artificial Neuronal Network (ANN) technique are investigated to classify the heart beats which are Multi Layer Perception (MLP) and Radial Basis Function (RBF). A calculation algorithm of the RBF centers is proposed. For the Atria Fibrillation anomalies, an artificial neural network was used as a pattern classifier to distinguish three classes of the cardiac arrhythmias. The different classes consist of the normal beats (N), the Arrhythmia (AFA) and Tachycardia (TFA) Atria Fibrillation cases. The global and the partition classifier are performed. The arrhythmias of MIT-BIH database are analyzed. The ANN inputs are the temporal and morphological parameters deduced from the electrocardiograph. Results. The simulation results illustrate the performances of the studied versions of the neural network and give the fault detection rate of the tested data, a rate of classification reaching the 3.7%. Conclusion. This system can constitute a mesh in a chain of automated diagnosis and can be a tool for assistance for the classification of the cardiac anomalies in the services of urgencies before the arrival of a qualified personal person. 196

This paper describes a novel approach to modelling a specific orthopaedic condition, Hallux Valgus; it is a complex deformity resulting in more than 140 possible surgical correction procedures, each focusing on different components of the deformity. Modelling it involves a level of complexity that cannot be readily tackled by techniques traditionally available in the medical domain. It was, therefore, appropriate for us to utilise machine learning techniques; namely through neural network detection and isolation, complemented by angle detection. We present results of running a machine learning algorithm to detect the deformity and end with recommendations as to how it may be utilised in judging successful surgical outcomes.

Educational data mininghas contributed to enhancing student academic performance by way of enabling stakeholders in academic institutions to have a pre-knowledge of the risks and dangers ahead and how to mitigate them. Prediction algorithms perform differently on dataset, and so, the need to develop models using different prediction algorithms and evaluating the result of such predictions is very important in order to be sure the best algorithm for a particular dataset is used.This work employed four classifiers: K-Nearest-Neighbour, Neural Network, Naïve Bayes and Decision Tree to model and, evaluated their models to know the performance of each on the target dataset. Their results were evaluated based on the various performance metrics. The results showed that Decision Tree had the highest accuracy on the dataset with test accuracy of 48.25% and therefore is the most suitable out of the four classifiers for performing prediction modelling on the dataset. Naïve Bayes is the second-best prediction model that can be used for predicting academic performance with an accuracy of 36.25%., followed by Neural Network with accuracy of 32.5 % and then K-Nearest Neighbour with accuracy of 32.5% but with lower precision, recall and area under Receiver Operating Characteristic curve.

Background: A vital tool for medical diagnosis, magnetic resonance imaging (MRI) produces high-resolution, non-ionizing images of inside body components. Despite its potential, MRI pictures frequently have noise and abnormalities that can mask important information, making a diagnosis difficult. Objective: This work uses a variety of cutting-edge machine learning approaches to improve the preprocessing and categorization of MRI data for more precise brain tumor identification. Methodologies: An integrated strategy was used, starting with the preprocessing of images using K-nearest neighbors (KNN) to lower noise and enhance quality. The photos were then graphed to enable more in-depth structural analysis, with nodes representing pixels and edges reflecting similar intensities. The photos were processed using Local Phase Quantization (LPQ) to extract textural information. A Multilayer Perceptron (MLP) was used for classification, and its hyperparameters were efficiently adjusted by Bayesian optimization. Findings: With 5,519 MRI pictures used for training and testing, the model achieved an accuracy of 87% on a test set of 1,104 images. The weighted average F1-score of 0.86 obtained from the classification data, along with noteworthy precision and recall values of 0.86 and 0.93, respectively, show how well the model identified brain tumors. Conclusion: The accuracy of brain tumor diagnosis from MRI images is greatly improved by the combination of KNN preprocessing, LPQ feature extraction, and MLP classification refined using Bayesian methods, highlighting the potential of integrated machine learning approaches in medical diagnostics.

Two popular representation learning paradigms are dictionary learning and deep learning. While dictionary learning focuses on learning ''basis'' and ''features'' by matrix factorization, deep learning focuses on extracting features via learning ''weights'' or ''filter'' in a greedy layer by layer fashion. This paper focuses on combining the concepts of these two paradigms by proposing deep dictionary learning and show how deeper architectures can be built using the layers of dictionary learning. The proposed technique is compared with other deep learning approaches, such as stacked autoencoder, deep belief network, and convolutional neural network. Experiments on benchmark data sets show that the proposed technique achieves higher classification and clustering accuracies. On a real-world problem of electrical appliance classification, we show that deep dictionary learning excels where others do not yield at-par performance. We postulate that the proposed formulation can pave the path for a new class of deep learning tools.

Two popular representation learning paradigms are dictionary learning and deep learning. While dictionary learning focuses on learning ''basis'' and ''features'' by matrix factorization, deep learning focuses on extracting features via learning ''weights'' or ''filter'' in a greedy layer by layer fashion. This paper focuses on combining the concepts of these two paradigms by proposing deep dictionary learning and show how deeper architectures can be built using the layers of dictionary learning. The proposed technique is compared with other deep learning approaches, such as stacked autoencoder, deep belief network, and convolutional neural network. Experiments on benchmark data sets show that the proposed technique achieves higher classification and clustering accuracies. On a real-world problem of electrical appliance classification, we show that deep dictionary learning excels where others do not yield at-par performance. We postulate that the proposed formulation can pave the path for a new class of deep learning tools.

In this work we propose a new deep learning tool called deep dictionary learning. Multi-level dictionaries are learnt in a greedy fashion, one layer at a time. This requires solving a simple (shallow) dictionary learning problem, the solution to this is well known. We apply the proposed technique on some benchmark deep learning datasets. We compare our results with other deep learning tools like stacked autoencoder and deep belief network; and state of the art supervised dictionary learning tools like discriminative KSVD and label consistent KSVD. Our method yields better results than all.

The implementation of machine learning in bioinformatics and medical analysis has brought about great improvement in the health care delivery sector of the economy. As the heart remains the pivotal organ of the human body pumping blood to all parts of the body the need for adequate care of this vital organ is required. This paper therefore reviews the different approach to electrical signal process of the heart through ECG analysis and proposed the development of an arrhythmia detection model using artificial neural network (ANN). The model developed in Python on Google Colab environment was trained and tested with dataset contains 3640 sample heart beats. The dataset was preprocessed and split in the ration of 70:30 for the training and the test dataset respectively and the model performance impressively well with a record 97.43 % accuracy showing the possibility of improvement after more training. The system indicates the average heartbeat of 72.0 and 12 peaks from the sample dataset showing no trace of arrhythmia.

Nowadays, electrical power system is considered as one of the most complicated artificial systems all over the globe, as social and economic development depends on intact, consistent, stable and economic functions. Owing to diverse random causes, accidental failures occur in electrical power systems. Considering this issue, this article aimed to propose the use of deep belief network (DBN) in detecting and classifying fault signals such as transient, sag and swell in the transmission line. Here, wavelet-decomposed fault signals are extracted and the fault is diagnosed based on the decomposed signal by the DBN model. Further, this article provides the performance analysis by determining the types I and II measures and root-mean-square-error (RMSE) measure. In the performance analysis, it compares the performance of the DBN model to various conventional models like linear support vector machine (SVM), quadratic SVM, radial basis function SVM, polynomial SVM, multilayer perceptron SVM,...

This article proposes a deep learning technique for the prevision of the geometric accuracy in single point incremental forming. Moreover, predicting geometric accuracy is one of the most crucial measures of part quality. Accordingly, roundness and positioning deviation are two indicators for measuring geometric accuracy and presenting two output variables. Two types of artificial intelligence learning approaches, that is, shallow learning and deep learning, are investigated and compared for forecasting geometrical accuracy in the single point incremental forming process. Therefore, the back-propagation neural network with one hidden layer is selected as the representative for shallow learning and deep belief network and stack autoencoder are chosen as the representatives for deep learning. Accurate prediction is closely related to the feature learning of single point incremental forming process parameters. The following six parameters were considered as input variables: sheet thick...

Deep reinforcement learning (RL) algorithms have recently achieved remarkable successes in various sequential decision making tasks, leveraging advances in methods for training large deep networks. However, these methods usually require large amounts of training data, which is often a big problem for real-world applications. One natural question to ask is whether learning good representations for states and using larger networks helps in learning better policies. In this paper, we try to study if increasing input dimensionality helps improve performance and sample efficiency of model-free deep RL algorithms. To do so, we propose an online feature extractor network (OFENet) that uses neural nets to produce good representations to be used as inputs to deep RL algorithms. Even though the high dimensionality of input is usually supposed to make learning of RL agents more difficult, we show that the RL agents in fact learn more efficiently with the high-dimensional representation than wi...

The rapid growth of streaming platforms has transformed the landscape of television and video consumption. Traditional television advertising, which relied on fixed schedules and generalized targeting, is increasingly being replaced by personalized and context-aware advertising strategies. This paper presents an AI-driven framework for real-time ad customization in streaming platforms, utilizing Generative Adversarial Networks (GANs), Reinforcement Learning (RL), and Long Short-Term Memory (LSTM) networks. These technologies enable dynamic ad content generation, optimal ad placement, and user behavior prediction, ensuring that advertisements are tailored to individual viewer preferences and interactions in real-time. Our experimental results show significant improvements in viewer engagement, prediction accuracy, and revenue generation, demonstrating the effectiveness of AI in transforming the ad experience. This personalized approach not only enhances user satisfaction but also maximizes ad revenue for content providers. The proposed framework provides a comprehensive solution to the challenges of delivering non-intrusive, highly engaging advertisements in a fastevolving streaming environment.

Significant advances in the automated glaucoma detection techniques have been made through the employment of the Machine Learning (ML) and Deep Learning (DL) methods, an overview of which will be provided in this paper. What sets the current literature review apart is its exclusive focus on the aforementioned techniques for glaucoma detection using the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines for filtering the selected papers. To achieve this, an advanced search was conducted in the Scopus database, specifically looking for research papers published in 2023, with the keywords "glaucoma detection", "machine learning", and "deep learning". Among the multiple found papers, the ones focusing on ML and DL techniques were selected. The best performance metrics obtained using ML recorded in the reviewed papers, were for the SVM, which achieved accuracies of 98.31%, 98.61%, 96.43%, 96.67%, 95.24%, and 98.60% in the ACRIMA, REFUGE, RIM-ONE, ORIGA-light, DRISHTI-GS, and sjchoi86-HRF databases, respectively, employing the REFUGE-trained model, while when deploying the ACRIMA-trained model, it attained accuracies of 98.92%, 99.06%, 98.27%, 97.10%, 96.97%, and 96.36%, in the same databases, respectively. The best performance metrics obtained utilizing DL recorded in the reviewed papers, were for the lightweight CNN, with an accuracy of 99.67% in the Diabetic Retinopathy (DR) and 96.5% in the Glaucoma (GL) databases. In the context of non-healthy screening, CNN achieved an accuracy of 99.03% when distinguishing between GL and DR cases. Finally, the best performance metrics were obtained using ensemble learning methods, which achieved an accuracy of 100%, specificity of 100%, and sensitivity of 100%. The current review offers valuable insights for clinicians and summarizes the recent techniques used by the ML and DL for glaucoma detection, including algorithms, databases, and evaluation criteria.

Background: Cloud computing, AI, and IoT technologies are revolutionizing healthcare by enabling predictive analytics and ongoing health monitoring. A hybrid technique that combines the Gray Wolf Optimization (GWO) algorithm with Deep Belief Networks (DBN) enhances disease prediction and real-time patient monitoring, particularly for chronic disorders. Objective: Designing and implementing a GWO-DBN hybrid model that improves predictive accuracy for chronic disease monitoring in cloud environments is the goal of this project. Utilizing wearable IoT devices and cloud capabilities, the goal is to combine scalable, realtime analysis for healthcare providers for reliable, remote illness management. Methods: Data collection, preprocessing, cloud storage, optimization, and monitoring are all integrated into the hybrid GWO-DBN architecture. The GWO algorithm chooses important features and optimizes DBN parameters. Cloud infrastructure enables real-time alerts, enabling prompt responses by healthcare providers based on ongoing patient health parameters. Results: The GWO-DBN model outperforms conventional methods with a 93% prediction accuracy, 90% sensitivity, and 95% specificity. A useful, high-performing solution for disease monitoring and resource optimization in the healthcare industry is offered by the cloud integration, that guarantees scalability and real-time notifications for healthcare providers. Conclusion: In real-time disease monitoring applications, the GWO-DBN model exhibits enhanced operational scalability, predictive accuracy, and efficiency. This cloud-based hybrid solution provides a viable model for proactive healthcare by enabling early diagnosis and resource allocation.

Deep belief network (DBN) is a probabilistic generative model with multiple layers of hidden nodes and a layer of visible nodes, where parameterizations between layers obey harmonium or restricted Boltzmann machines (RBMs). In this paper we present restricted deep belief network (RDBN) for multi-view learning, where each layer of hidden nodes is composed of view-specific and shared hidden nodes, in order to learn individual and shared hidden spaces from multiple views of data. View-specific hidden nodes are connected to corresponding view-specific hidden nodes in the lower-layer or visible nodes involving a specific view, whereas shared hidden nodes follow inter-layer connections without restrictions as in standard DBNs. RDBN is trained using layer-wise contrastive divergence learning. Numerical experiments on synthetic and real-world datasets demonstrate the useful behavior of the RDBN, compared to the multi-wing harmonium (MWH) which is a two-layer undirected model.

Network Intrusion Detection (ID) attempts to detect diverse security attacks using a security system to monitor, analyze, detect, and respond to threats. In Software Defined Networking (SDN), ID that typically occurs at the controller is influenced by application policies and is advantageous over generic networking due to the network and flow awareness of the controller and the threat response capability of the controller itself. A blockchain system embodies a succession of attached blocks that implicitly maintain the purity, preserve the indisputability, and preserve the quasi-anonymity of its transactions/activities because of consensus protocols and cryptographic algorithms. Being the front-line reviewers for BCcentered ID in SDN, we group blockchain roles in ID in three ways and elaborately examine those pertaining to ID techniques and ID approaches, blockchain components, attack detection, network components, and so forth. We stacked a preliminary sample of 97 paper sources by sifting through the articles for selection guidelines inquired from virtual libraries, applying an elaborate and overtime procedure. Centered upon this examination, blockchain-centered ID in SDN involves collaborative ID by engaging smart contracts while storing data required for threat detection in blockchain, off-chain ID while applying blockchain for trust management, and applying blockchain for authentication and secure data storage for off-chain ID. Elaborate examination divulges that from blockchain-centered SDN ID proposals, 48.9% apply blockchaincentered secure storage for off-chain ID, 91.2% engage systematic blockchain architecture, 8.9% apply PoAuthority consensus, 100% online, 75.5% active response, 71.1% fully distributed, and 57.8% apply hybrid host and networkcentered ID, with DDoS being the most dominant (17.8%) specific attack category. Finally, we explore the potentials and hardships of the schema of blockchain-centered ID in SDN and then supply insights to overtake them.

In recent years multilayer perceptrons (MLPs) with many hidden layers Deep Neural Network (DNN) has performed surprisingly well in many speech tasks, i.e. speech recognition, speaker verification, speech synthesis etc. Although in the context of F 0 modeling these techniques has not been exploited properly. In this paper, Deep Belief Network (DBN), a class of DNN family has been employed and applied to model the F 0 contour of synthesized speech which was generated by HMM-based speech synthesis system. The experiment was done on Bengali language. Several DBN-DNN architectures ranging from four to seven hidden layers and up to 200 hidden units per hidden layer was presented and evaluated. The results were compared against clustering tree techniques popularly found in statistical parametric speech synthesis. We show that from textual inputs DBN-DNN learns a high level structure which in turn improves F 0 contour in terms of objective and subjective tests.

Many patients' lives are being saved by image-guided interventions, and the image registration issue must be regarded as the most difficult and complex problem to solve. However, the latest enormous advancements in machine learning (ML), which include the potential to deploy deep neural networks (DNNs) on modern many-core GPUs, have created a promising opportunity for tackling a variety of medical applications, including registration. The most recent research on medical image registration with DNNs is reviewed in detail in the presented work. All of the relevant papers that have already been published in the subject are included in the systematic review. This thorough overview includes a detailed discussion as well as survey of important ideas, statistical analysis from many perspectives, novelties and key contributions, confiding challenges, future directions, key-enabling approaches, and prospective trends. For readers who are actively involved in the subject, researching state-of-the-art and hoping to present a contribution to future publications, the presented study offers a deep grasp and insight.

The Semi-Parallel Deep Neural Network (SPDNN) idea is explained in this article and it has been shown that the convergence of the mixed network is very close to the best network in the set and the generalization of SPDNN is better than all the parent networks.

Nowadays, big data is directing the entire advanced world with its function and applications. Moreover, to make better decisions from the ever emerging big data belonging to the respective organizations, deep learning (DL) models are required. DL is also widely used in the sentiment classification tasks considering data from social networks. Furthermore, sentiment classification signifies the best way to analyze the big data and make decisions accordingly. Analyzing the sentiments from big data applications is quite challenging task and also requires more time for the execution process. Therefore, to analyze and classify big data emerging from social networks in a better way, DL models are utilized. DL techniques are being used among the researchers to get high end results. A novel Ant Colony-based Deep Belief Neural Network (AC-DBN) framework is proposed in this research. Drug review tweets are opted to perform sentiment classification by using the proposed framework in python envi...

This paper proposes a high-resolution non-volatile memory cell design that addresses the most substantial limitations associated with the effective implementation of analog long-term memory storage solution. Prior research efforts often suffer from limited resolution, hindering their ability to accurately represent fine-grained weight adjustments required for effective learning in analog neuromorphic systems. This work effort has been channeled toward crafting conductive circuit designs using 90 nm complementary metal-oxide semiconductor technology for on-chip learning applications in analog neuromorphic systems. The operational mechanism of the cell involves the storage of charge on the floating gate of the NM0 transistor. The writing process is accomplished through hot-electron injection, while the erasure of stored information is executed via gate oxide tunneling. An advantageous feature of this cell is its capability to facilitate simultaneous reading and writing of data. The reduction of errors that may arise due to oxide mismatch or charge trapping is achieved through feedback control incorporation during the writing phase. The memory reveals clear synaptic behavior characteristics in storing and retrieving analog information reliably including, good memory cell resolution, good charge retention rate, reliable operation in noisy environments, and high resolution with faster learning with a power consumption of 1.06 µW and an output current of 10 µA under a typical operating voltage of 1 V. This strategic implementation enhances precise and reliable weight updates within neuromorphic analog artificial neural networks, which is essential for ensuring accurate on-chip learning outcomes as well as minimizing power consumption.

Nowadays, spam detection has been one of the foremost machine learning-oriented applications in the context of security in computer networks. In this work, we propose to learn intrinsic properties of e-mail messages by means of Restricted Boltzmann Machines (RBMs) in order to identity whether such messages contain relevant (ham) or non-relevant (spam) content. The main idea contribution of this work is to employ Harmony Search-based optimization techniques to fine-tune RBM parameters, as well as to evaluate their robustness in the context spam detection. The unsupervised learned features are then used to feed the Optimum-Path Forest classifier, being the original features extracted from e-mail content and compared against the new ones. The results have shown RBMs are suitable to learn features from e-mail data, since they obtained favorable results in the datasets considered in this work.

In this article, the regularity of the global solutions to atmospheric circulation equations with humidity effect is considered. Firstly, the formula of the global solutions is obtained by using the theory of linear operator semigroups. Secondly, the regularity of the global solutions to atmospheric circulation equations is presented by using mathematical induction and regularity estimates for the linear semigroups.

Energy theft is a significant problem that needs to be addressed for effective energy management in smart cities. Smart meters are highly utilized in smart cities that help in monitoring the energy utilization level and provide information to the users. However, it is not able to detect energy theft or over-usage. Therefore, we have proposed a multi-objective diagnosing structure named an Energy Theft Prevention System (ETPS) to detect energy theft. The proposed system utilizes a combination of machine learning techniques Gated Recurrent Unit (GRU), Grey Wolf Optimization (GWO), Deep Recurrent Convolutional Neural Network (DDRCNN), and Long Short-Term Memory (LSTM). The statistical validation has been performed using the simple moving average (SMA) method. The results obtained from the simulation have been compared with the existing technique in terms of delivery ratio, throughput, delay, overhead, energy conversation, and network lifetime. The result shows that the proposed system is more effective than existing systems.