Neural Network Research Articles

Real-Time UAV Recognition Through Advanced Machine Learning for Enhanced Military Surveillance

In an era where the military utilization of Unmanned Aerial Vehicles (UAVs) has become essential for surveillance and operational operations, our study tackles the growing demand for real-time, accurate UAV recognition. The rise of UAVs presents numerous safety hazards, requiring systems that distinguish UAVs from non-threatening phenomena, such as birds. This research study conducts a comparative examination of advanced machine learning models, aiming to address the challenge of real-time aerial classification in diverse environmental conditions without model retraining. This research employs extensive datasets to train and validate models such as Neural Networks, Support Vector Machines, ensemble methods, and Gradient Boosting Machines. The fashions are evaluated based on accuracy, forgetfulness, and processing efficiency—criteria determining the viability of real-time operational scenarios. The findings indicate that Neural Networks exhibit enhanced performance, demonstrating exceptional accuracy in distinguishing UAVs from birds. This culminates in our primary assertion: Neural Networks possess vital operational security ramifications and can markedly enhance the allocation of defense resources. The findings significantly improve surveillance systems, highlighting the effectiveness of machine-learning methods in real-time UAV identification. Moreover, incorporating Neural Network systems into military defenses is recommended to enhance decision-making capabilities and security operations. Foresee forthcoming UAV developments and advocate for regular model updates to keep up with increasingly nimble and perhaps stealthier drone designs.

The critical avalanche of an excitation–inhibition neural network composed of Izhikevich neurons is studied based on the bifurcation of the mean-field

In recent years, there has been significant interest in exploring how the brain efficiently propagates and processes information. When the system is in a critical state, it may have preferable information transmission, calculation, and processing abilities, so the study of this state has attracted wide attention. Here, we primarily focus on the excitation–inhibition (E–I) neural network composed of Izhikevich neurons and the derived accurate mean-field model. To understand the critical neuronal avalanche more comprehensively and deeply, we consider the spike characteristics of individual neurons (spike or burst firing), the criticality of neural networks, and the bifurcations of the mean-field. It is observed that no matter the spike or burst firing, near different bifurcations of the mean-field, the neural network appears in phase transition and the neuronal avalanche satisfies the critical condition. In other words, the bifurcation of this mean-field model can more accurately and conveniently predict the occurrence of critical avalanches, thus narrowing the detection range. These findings contribute to a deeper understanding of critical avalanches. By establishing the relationship between the spike of individual neurons, microscopic neural network, and macroscopic mean-field, new insight is provided into the dynamic origin of critical neuronal avalanches.

The improved BW model are optimized based on MLP neural network

<sec>The nuclear mass model has significant applications in nuclear physics, astrophysics, and nuclear engineering. The accurately predictions of binding energy are crucial for research on nuclear structure, reactions, and decay. However, traditional mass models exhibit large errors in double magic number regions and heavy nuclei regions. These models struggle to effectively describe shell effects and parity effects in nuclear structures, and also fail to capture the subtle differences observed in experimental results.</sec><sec>This study shows the strong modeling capabilities of MLP neural networks, which optimizes the parameters of the nuclear mass model, and reduces prediction errors in key regions and globally. The neural network takes the features as neutron number, proton number, and binding energy, and the labels as mass-model coefficients. The training set is the multiple sets of calculated nuclear mass model coefficients. Through extensive experimentation, the optimal parameters are determined to ensure model convergence speed and stability. The Adam optimizer is employed to adjust the weights and biases of the network, for reducing the mean squared error loss during training.</sec><sec>The trained neural network model with minimal loss was used to predict the optimal coefficients of the nuclear mass model based on the AME2020 dataset. The optimized BW2 model significantly reduces root-mean-square errors in double magic number and heavy nuclei regions. Specifically, the optimized model achieved reductions of approximately 28%, 12%, and 18% in root-mean-square errors near Z(N) = 50, Z(N) = 50 (82), and Z(N) = 82 (126), respectively. In heavy nuclei regions, the errors were reduced by 48%. The BW3 model, incorporating higher-order symmetry energy terms, reduced global root-mean-square errors from 1.86 MeV to 1.63 MeV after parameter optimization using the neural network.</sec><sec>This work reveals that the model with newly optimized coefficients not only exhibits significant error reductions near double magic numbers but also shows improvements in binding energy predictions for both neutron-rich and neutron-deficient nuclei. Furthermore, the model demonstrates good improvements in describing parity effects, accurately capturing parity-related differences in isotopic chains with varying proton numbers.</sec><sec>This study demonstrates the tremendous potential of MLP neural networks in optimizing nuclear mass model parameters and provides a novel method for optimizing parameters in more complex nuclear mass models. Moreover, the proposed method applies to nuclear mass models with implicit or nonlinear relationships, offering new perspectives for further development of nuclear mass models.</sec>

Artificial Intelligence-Based Direct Power Control for Power Quality Improvement in a WT-DFIG System via Neural Networks: Prediction and Classification Techniques

This paper discusses the improvement of power quality injected into the AC grid. This approach is achieved by enhancing the quality of injected power signals and mastering the active and reactive power exchanged between the DFIG based wind turbine (WT-DFIG) and the electrical grid, resulting in an improvement of the overall system performance and efficiency. This study includes all WT-DFIG operating modes, successively and continuously, as well as all local reactive power compensation modes. Therefore, novel control strategies are proposed in this paper for wind energy conversion systems based on artificial intelligence techniques. These techniques include Neural Network Prediction (PNN-DPC) and Classification (CNN-DPC). They aim to eliminate the drawbacks and difficulties associated with conventional Direct Power Control (C-DPC), while retaining its advantages. The paper also provides a thorough explanation of the mathematical models for neural network techniques and WT-DFIG system models. The MATLAB/Simulink environment is used to investigate the performance of the proposed techniques under different conditions and operating modes related to different scenarios. The results reveal a significant reduction in the ripple of the generated active power and the compensated local reactive power, better quality of the generated signal currents and a remarkable reduction in the current total harmonic distortion (THD). Furthermore, compared to C-DPC, PNN-DPC achieves a reduction of 72.07% in active power ripples, 77.07% in reactive power ripples, and 76.79% in current Total Harmonic Distortion (THD). CNN-DPC shows similar improvements with 72.04%, 77.13%, and 76.54% of reductions respectively. In addition, CNN-DPC slightly outperforms PNN-DPC. Nevertheless, both proposed control techniques show significant improvements in all characteristics compared to other methods. Consequently, the proposed control strategies indicate that artificial intelligence has the potential to improve the power quality and performance of wind power conversion system.

Comparison of Classical, Neural Network and Hybrid Models for Hysteretic Single-tendon Catheter Kinematics.

While robotic control of catheter motion can improve tip positioning accuracy, hysteresis arising from tendon friction and flexural deformation degrades kinematic modeling accuracy. In this paper, we compare the capabilities of three types of models for representing the forward and inverse kinematic maps of a clinical single-tendon cardiac catheter. Classical hysteresis models, neural networks and hybrid combinations of the two are included. Our results show that modeling accuracy is best when models are trained using motions corresponding to the anticipated clinical motions. For sinusoidal motions, recurrent neural network models provide the best performance. For point-to-point motions, however, a simple backlash model can provide comparable performance to a recurrent neural network.

Superior Classification of Brain Cancer Types Through Machine Learning Techniques Applied to Magnetic Resonance Imaging

Brain cancer remains one of the most challenging medical conditions due to its intricate nature and the critical functions of the brain. Effective diagnostic and treatment strategies are essential, particularly given the high stakes involved in early detection. Magnetic Resonance (MR) imaging has emerged as a crucial modality for the identification and monitoring of brain tumors, offering detailed insights into tumor morphology and behavior. Recent advancements in artificial intelligence (AI) and machine learning (ML) have revolutionized the analysis of medical imaging, significantly enhancing diagnostic precision and efficiency. This study classifies three primary brain tumor types—glioma, meningioma, and general brain tumors—utilizing a comprehensive dataset comprising 15,000 MR images obtained from Kaggle. We evaluated the performance of six distinct machine learning models: K-Nearest Neighbors (KNN), Neural Networks, Logistic Regression, Support Vector Machine (SVM), Decision Trees, and Random Forests. Each model's effectiveness was assessed through multiple metrics, including classification accuracy (CA), Area Under the Curve (AUC), F1 score, precision, and recall. Our findings reveal that KNN and Neural Networks achieved remarkable classification accuracies of 98.5% and 98.4%, respectively, significantly surpassing the performance of other evaluated models. These results underscore the promise of ML algorithms, particularly KNN and Neural Networks, in improving the diagnostic process for brain cancer through MR imaging. Future research will focus on validating these models with real-world clinical data, aiming to refine and enhance diagnostic methodologies, thus contributing to the development of more accurate, efficient, and accessible tools for brain cancer diagnosis and management.

A Cost-Minimized Task Migration Assignment Mechanism in Blockchain Based Edge Computing System

Background: Cloud computing is usually introduced to execute computing intensive tasks for data processing and data mining. As a supplement to cloud computing, edge computing is provided as a new paradigm to effectively reduce processing latency, energy consumption cost and bandwidth consumption for time-sensitive tasks or resource-sensitive tasks. To better meet such requirements during task assignment in edge computing systems, an intelligent task migration assignment mechanism based on blockchain is proposed, which jointly considers the factors of resource allocation, resource control and credit degree. Methods: In this paper, an optimization problem is firstly constructed to minimize the total cost of completing all tasks under constraints of delay, energy consumption, communication, and credit degree. Here, the terminal node mines computing resources from edge nodes to complete task migration. An incentive method based on blockchain is provided to mobilize the activity of terminal nodes and edge nodes, and to ensure the security of the transaction during migration. The designed allocation rules ensure the fairness of rewards for successfully mining resource. To solve the optimization problem, an intelligent migration algorithm that utilizes a dual “actor-reviewer” neural network on inverse gradient update is proposed which makes the training process more stable and easier to converge. Results: Compared to the existing two benchmark mechanisms, the extensive simulation results indicate that the proposed mechanism based on neural network can converge at a faster speed and achieve the minimal total cost. Conclusion: To satisfy the requirements of delay and energy consumption for computing intensive tasks in edge computing scenarios, an intelligent, blockchain based task migration assignment mechanism with joint resource allocation and control is proposed. To realize this mechanism effectively, a dual “actor-reviewer” neural network algorithm is designed and executed.

How the technologies behind self-driving cars, social networks, ChatGPT, and DALL-E2 are changing structural biology.

The performance of deep Neural Networks (NNs) in the text (ChatGPT) and image (DALL-E2) domains has attracted worldwide attention. Convolutional NNs (CNNs), Large Language Models (LLMs), Denoising Diffusion Probabilistic Models (DDPMs)/Noise Conditional Score Networks (NCSNs), and Graph NNs (GNNs) have impacted computer vision, language editing and translation, automated conversation, image generation, and social network management. Proteins can be viewed as texts written with the alphabet of amino acids, as images, or as graphs of interacting residues. Each of these perspectives suggests the use of tools from a different area of deep learning for protein structural biology. Here, I review how CNNs, LLMs, DDPMs/NCSNs, and GNNs have led to major advances in protein structure prediction, inverse folding, protein design, and small molecule design. This review is primarily intended as a deep learning primer for practicing experimental structural biologists. However, extensive references to the deep learning literature should also make it relevant to readers who have a background in machine learning, physics or statistics, and an interest in protein structural biology.

Fusion Model of Quantum Wavelet Transform and Neural Network for Video Coding on the Internet of Things Environment

Solving the video compression problem requires a multi-faceted approach, balancing quality, efficiency, and computational demands. By leveraging advancements in technology and adapting to the evolving needs of video applications, it is possible to develop compression methods that meet the challenges of the present and future digital landscape. To address these objectives, machine learning and AI approaches can be utilized to predict and remove redundancies more effectively, optimizing compression algorithms dynamically based on content. Still, state-of-the art neural network-based video compression models need large and diverse datasets to generalize well across different types of video content. Wavelets can provide both time (spatial) and frequency localization, making them highly effective for video compression. This dual localization allows wavelet transforms to handle both rapid changes in video content and slow-moving scenes efficiently, leading to better compression ratios. Yet, some wavelet coefficients may be more critical for maintaining visual quality than others. Inaccurate quantization can lead to noticeable degradation. For the first time, the suggested model combine Quantum Wavelet Transform (QWT) and Neural Networks (NN) for video compression. This fusion model aims to achieve higher compression ratios, maintain video quality, and reduce computational complexity by utilizing QWT’s efficient data representation and NN’s powerful pattern recognition and predictive capabilities. Quantum bits (qubits) can encode large amounts of information in their quantum states, enabling more efficient data representation. This is especially useful for encoding large video files. Furthermore, quantum entanglement allows for correlated data representation across qubits, which can be exploited to capture intricate details and redundancies in video data more effectively than classical methods. The experimental results reveal that QWT achieves a compression ratio of almost twice that of traditional WT for the same video, maintaining superior visual quality due to more efficient redundancy elimination.

Fixed-time synchronization of coupled neural networks under Hölder continuous nonlinear activation function

This paper investigates the fixed-time stability of dynamical systems and its application on the fixed-time synchronization (FTSYN) of the coupled neural networks (NNs). Firstly, a new theorem on fixed-time stability is established, along with an estimation of the settling time. Based on the new theorem and Lyapunov direct method, two different sufficient criteria are derived to ensure that the NNs in presence of Hölder continuity can achieve synchronization within fixed-time based on 1-norm and 2-norm, separately. Finally, two numerical simulations are conducted to demonstrate the validity and accuracy of the main findings.

A deep neural network-based multi-layer classifier ensembles for intrusion detection in fog-based Internet of Things environments

Intrusion Detection Systems (IDS) are regarded as an efficient security mechanism in the Internet of Things (IoT). IDSs in the IoT face critical challenges, such as a vast amount of data with many features, limited resources, detection of unknown attacks, and many false alarms. To overcome these issues, this paper proposes a novel framework, namely a Multi-Layer Multi-Classification System (MLMCS). The proposed framework consists of four steps: In the first step, the data collected from information sources are pre-processed. The second step divides the feature space into several subspaces to reduce the model's complexity. The third step proposes a multilayer base classifier called a Multi-Layer Classifier System (MLCS) for each feature subspace. Each MLCS is layered to transform a multi-class classification into several binary classifications so that a specific category of attacks has been detected in each layer based on the Group Method of Data Handling (GMDH) neural network. In the fourth step, the results obtained from MLCSs based on the Dempster-Shafer theory are combined to reduce the problems related to overfitting. The simulation results were gained using three datasets, i.e., NSL-KDD, UNSW-NB15, and TON-IoT. They indicated that the proposed method, compared with other methods, improved F1-score by 5.64%, 0.65%, and 0.11%, respectively. Furthermore, the false alarm rate was reduced by 2.66%, 2.52%, and 0.02%. Moreover, the impact of four well-known adversarial attacks on the dataset was investigated. It was observed that the MLMCS method, with an average 5.21% reduction, could perform well against these attacks.

Reduced basis approximations of parameterized dynamical partial differential equations via neural networks

Reduced basis approximations of parameterized dynamical partial differential equations via neural networks

Green behavior propagation analysis based on statistical theory and intelligent algorithm in data-driven environment.

The correlation between green behavior and energy efficiency is growing due to the heightened focus on energy efficiency among individuals. This paper introduces a three-layer network model to analyze the relationships among information diffusion, awareness and green behavior spreading. We have analyzed the probability tree of state transfer across 12 states by using Microscopic Markov Chain Approach (MMCA) and derived the state transfer equations for each state to compute the state transition threshold. In addition, we use the reaction-diffusion system to model the interaction between space and time changes for each state in the green behavior propagation layer. The equilibrium point of the system is defined, and the criteria for Turing bifurcation are identified. The optimal control approach achieves parameter identification, and this study validates the theory through several numerical simulations. Meanwhile, the effectiveness of parameter identification based on the convolutional neural network (CNN) and optimal control is compared. The data on China's electrical energy generation is predicted and compared by using three neural networks and an autoregressive integrated moving average (ARIMA) model. Further, considering clean energy generation as a green behavior, we fit the data on the percentage of clean energy generation by applying a Microscopic Markov Chain model and a reaction-diffusion system.

Distributed Finite Memory Online Learning Strategy for Multi-UAV Systems With Neural Networks

Distributed Finite Memory Online Learning Strategy for Multi-UAV Systems With Neural Networks

On geometry of fixed figures via φ−interpolative contractions and application of activation functions in neural networks and machine learning models

On geometry of fixed figures via φ−interpolative contractions and application of activation functions in neural networks and machine learning models

View-symmetric representations of faces in human and artificial neural networks.

View symmetry has been suggested to be an important intermediate representation between view-specific and view-invariant representations of faces in the human brain. Here, we compared view-symmetry in humans and a deep convolutional neural network (DCNN) trained to recognise faces. First, we compared the output of the DCNN to head rotations in yaw (left-right), pitch (up-down) and roll (in-plane rotation). For yaw, an initial view-specific representation was evident in the convolutional layers, but a view-symmetric representation emerged in the fully-connected layers. Consistent with a role in the recognition of faces, we found that view-symmetric responses to yaw were greater for same identity compared to different identity faces. In contrast, we did not find a similar transition from view-specific to view-symmetric representations in the DCNN for either pitch or roll. These findings suggest that view-symmetry emerges when opposite rotations of the head lead to mirror images. Next, we compared the view-symmetric patterns of response to yaw in the DCNN with corresponding behavioural and neural responses in humans. We found that responses in the fully-connected layers of the DCNN correlated with judgements of perceptual similarity and with the responses of higher visual regions. These findings suggest that view-symmetric representations may be computationally efficient way to represent faces in humans and artificial neural networks for the recognition of identity.

Enhancing Object Detection and Classification Using White Shark Optimization with Deep Learning on Remote Sensing Images

Remote sensing (RS) object detection is extensively applied in the fields of civilian and military. The important role of remote sensing is to identify objects like planes, ships, harbours airports, etc., and then it can attain position information and object classification. It is of considerable importance to use RS images for observing the densely organized and directional objects namely ships and cars parked in harbours and parking areas. The object detection (OD) process involves object localization and classification. Due to its wide coverage and longer shooting distance, Remote sensing images (RSIs) have hundreds of smaller objects and dense scenes. Deep learning (DL), in particular convolution neural network (CNN), has revolutionized OD in different fields. CNN is devised to automatically learn the hierarchical representation of data, which makes them fit for feature extraction. Hence, the study proposes a new white shark optimizer with DL-based object detection and classification on RSI (WSODL-ODCRSI) method. The purpose of the WSODL-ODCRSI model is to classify and detect the presence of the objects in the RSI. To accomplish this, the WSODL-ODCRSI model uses a modified single-shot multi-box detector (MSSD) for the OD process. The next stage of OD is the object classification process, which takes place with the use of the Elman Neural Network (ENN) algorithm. The WSO algorithm is exploited as a parameter-tuning model for improving the object classification results of the ENN approach. The stimulated study of the WSODL-ODCRSI algorithm has been established on the benchmark data set and the outcomes underlined the promising performance of the WSODL-ODCRSI model on the object process of classification

Predictive crystallography at scale: mapping, validating, and learning from 1000 crystal energy landscapes.

Computational crystal structure prediction (CSP) is an increasingly powerful technique in materials discovery, due to its ability to reveal trends and permit insight across the possibility space of crystal structures of a candidate molecule, beyond simply the observed structure(s). In this work, we demonstrate the reliability and scalability of CSP methods for small, rigid organic molecules by performing in-depth CSP investigations for over 1000 such compounds, the largest survey of its kind to-date. We show that this highly-efficient force-field-based CSP approach is superbly predictive, locating 99.4% of observed experimental structures, and ranking a large majority of these (74%) as among the most stable possible structures (to within uncertainty due to thermal effects). We present two examples of insights such large predicted datasets can permit, examining the space group preferences of organic molecular crystals and rationalising empirical rules concerning the spontaneous resolution of chiral molecules. Finally, we exploit this large and diverse dataset for developing transferable machine-learned energy potentials for the organic solid state, training a neural network lattice energy correction to force field energies that offers substantial improvements to the already impressive energy rankings, and a MACE equivariant message-passing neural network for crystal structure re-optimisation. We conclude that the excellent performance and reliability of the CSP workflow enables the creation of very large datasets of broad utility and explanatory power in materials design.

Fractal-domain Transformer Based on Learnable Multifractal Spectrum for Chaotic Systems Classification

Conventional deep learning in the spatiotemporal-frequency domain frequently encounter challenges in terms of slow convergence rates and limited generalization, particularly for classification of chaotic systems. To address these limitations, this paper introduces a novel fractal-inspired deep network model, specifically, the Multifractal Spectrum Transformer (MFS-Transformer), grounded in learnable multifractal analysis. Initially, we put forward the conceptual framework of fractal learning, compared with traditional fractal signal processing methodologies and spatial-temporal domain learning paradigms. Subsequently, a Learnable Multifractal Spectrum (LMFS) derived from 3D spatial gridding, coupled with fractal-domain filtering, is proposed to construct the iterative learning process within the fractal domain. Further, we formulate the MFS-Transformer, an innovative architecture that integrates multi-channel embedding, LMFS, fractal-domain filtering, residual fusion mechanisms, a mixer module, and a classifier, tailored for chaotic system classification. Ultimately, we evaluate the efficacy of our model in classifying 3D chaotic systems under stringent conditions of short-term sequences and low Signal-to-Noise Ratio (SNR). Experimental outcomes underscore the substantial performance gains achieved by the MFS-Transformer, with classification accuracy enhancements of 13.34% and 5.00% over existing Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), respectively, under SNR=0dB and 32-sample sequences. These findings validate the superiority of the MFS-Transformer in addressing the complexities of chaotic system classification under complex scenarios. This research not only advances the frontier of fractal deep learning but also presents a novel perspective and methodology for tackling intricate spatiotemporal classification problems.

Effect of model architecture and input parameters to improve performance of artificial intelligence models for estimating concrete strength using SonReb

The use of Artificial Intelligence (AI) with the non-intrusive SonReb method, which combines Ultrasonic Pulse Velocity (UPV) and Rebound Number (RN) to predict concrete compressive strength, has attracted increasing attention in recent years. This study introduces a novel approach to improve AI models for predicting concrete strength, making them more suitable for future practical applications. One of the key challenge in AI models is the number of input parameters; while more inputs often improve accuracy, they are typically impractical for most applications dealing with existing structures (e.g., requiring detailed concrete mix design information that is often unavailable). SonReb AI-based models which use only two input parameters (UPV and RN) have shown reasonable accuracy, but their general use is limited by adoption of different testing standards which precludes the development of large databases. This study aims to improve two-parameter SonReb-based AI models through the addition of a binary input variable that represents the type of the specimen geometry (cube or cylinder) and investigates the effect of model architecture by comparing three different AI algorithms: Artificial Neural Networks (ANN), Deep Neural Networks (DNN), and Adaptive Neuro-Fuzzy Inference Systems (ANFIS). Six AI models were developed using 514 data points from experimental tests and collected data, and an unbiased data splitting method was applied for training and testing. The results showed that including specimen geometry improved model accuracy across all AI algorithms. The results of this study show that regardless of AI architecture, the proposed novel approach not only improves the accuracy of models, but also enables the use of larger databases containing both cubic and cylindrical specimens.