Computational Intelligence最新文献

Unlike traditional systems, cloud infrastructures are dynamic and distributed, necessitating specialized methods to handle distributed denial of service (DDoS) attacks that flood services with traffic, causing downtime or performance issues. This research introduces a novel approach for sensing and preventing Cloud DDoS attacks by addressing the unique challenges of dynamic and distributed cloud environments. The novelty of this work lies in three main contributions: (i) Preprocessing, (ii) Feature extraction, (iii) Attack detection, and (iv) Attack mitigation. In the preprocessing phase, the input dataset is balanced using random undersampling and processed with an enhanced decimal scaling (EDS) normalization technique. The feature extraction phase involves retrieving enhanced correntropy-based features, higher-order statistical (HOS) features, and raw features. Attack detection is carried out using a novel deep learning (DL)-based hybrid model named improved LinkNet-PolyNet (ILPNet), which combines improved LinkNet (ILNet) and PolyNet. This approach offers improved accuracy over the conventional LinkNet model by addressing bottlenecks between the encoder and decoder blocks. Lastly, an enhanced entropy-based mitigation approach is used to reduce detected attacks. The efficiency of this approach is evaluated through comparisons with existing methods using detailed performance, statistical, and ablation analyses. The ILPNet strategy demonstrated superior performance compared to conventional methods, achieving an accuracy of 0.960, a precision of 0.953, and an F-measure of 0.961.

Lymphoma is commonly considered as cancer that affects the entire organs of the body. The accurate classification of malignant lymphomas is helpful for providing better treatment plans to patients. Usually, the lymphoma types are differentiated by cytologic features and growth patterns. Moreover, the abnormal cell variations are effectively identified through the immunologic, genetic, and clinical features that are useful in making the diagnosis. The gap between pattern analysis and cancer diagnostics is effectively handled with the development of computer vision methods. Computed tomography (CT)-based image analysis and Positron Emission Tomography (PET)-based image analysis for the classification of malignant lymphomas have some disadvantages, including a lack of inter and intra-observer variability. Thus, a robust deep learning-aided malignant lymphoma classification model is implemented to identify the specific type of lymphoma. Initially, the input images are acquired from benchmark databases. The required images are processed via the proposed Hybrid Adaptive and Attentive Networks (HAAN) for the malignant lymphoma classification. Therefore, it is the combination of Dilated MobilenetV2 with Convolutional-Recurrent Neural Network (RNN) for the specific cancer subtype classification process. The functionality of the suggested network is enhanced by tuning the parameters in the network via the Revised Iteration-based Peregrine Falcon Optimization (RIPFO). Thus, the proposed model processes large volumes of input images quickly and accurately for obtaining efficient malignant lymphoma classification results. Finally, the performance of the developed framework is estimated with conventional methods. The analysis results prove that the accuracy of the malignant lymphoma subtype classification framework is higher than the baseline classification mechanisms. In order to prove the model effectiveness, the accuracy of the developed technique shows 93.67%, 94.74%, and 95.36% in terms of diverse activation functions like linear, sigmoid, and ReLU, respectively.

Agriculture plays a fundamental role in the global economy, contributing about 6% to global gross domestic product, while in many low-income nations, it accounts for up to 25%–40% of gross domestic product. Plants provide around 75% of the food consumed by humans and are the primary nutritional source for animals, making plant health crucial to global food security. However, plant diseases represent a serious threat to agricultural productivity, causing 10%–30% loss in global crop yields annually. The early detection and diagnosis of such diseases are thus vital to sustaining agricultural productivity. While deep learning techniques—particularly deep clustering—have shown promise in image-based disease detection, they often face limitations when dealing with complex and multi-modal data. To overcome these limitations, we propose an enhanced Dynamic Autoencoder method based on a Gaussian Mixture Model for plant disease detection. Unlike conventional methods that rely heavily on labeled data, our approach leverages probabilistic clustering and adaptive feature learning to distinguish between healthy and diseased plants more effectively. We evaluate our method using the PlantVillage dataset and demonstrate that our approach outperforms traditional deep learning and clustering approaches. The experimental results highlight the model's superior accuracy and normalized mutual information, underscoring its potential for scalable and intelligent plant disease monitoring in smart agriculture.

Continuum robot is a type of robot with continuous flexible skeleton, which plays an important role in production and life. At present, the position and orientation control (POC) scheme is not widely implemented in continuum robots, and it does not pay attention to the dynamic tracking error of the end-effector orientation, which has obvious limitations for completing the task. In view of the shortcomings of the existing research, this paper proposes a hybrid neural network-based algorithm for controlling the position and orientation of a continuum robot. In this algorithm, the POC scheme is applied to the continuum robot. Based on the kinematics equation of the continuum robot, the corresponding dynamic neural network (DNN) and the dynamic tracking algorithm of the end-effector orientation are obtained by theoretical derivation. The simulation results show that the proposed scheme reduces the position error of the end-effector of the continuum robot and improves operability. Specifically, the tracking error of the Euler angle converges quickly, and the ideal Euler angle is basically consistent with the actual Euler angle. The comparison between the experimental results and the existing control methods verifies the feasibility and advancement of the scheme.

Harmonic noise interference is commonly encountered during the operation of robotic manipulators and is often accompanied by various types of non-harmonic noise (e.g., polynomial, constant, and random noise). To tackle the challenges presented by these mixed-harmonic noise environments, this paper proposes a multi-noise-resistance low-computational-complexity zeroing neural network (MNR-LCCZNN) model. The proposed framework integrates an adaptive compensation mechanism to effectively suppress harmonic noise and employs a low-computational-complexity zeroing neural network (LCCZNN) structure that eliminates the need for matrix inversion, thereby enabling efficient handling of multi-task constraints. Furthermore, the incorporation of an advanced activation function significantly improves the model's convergence speed and robustness under noise-mixture conditions. Theoretical analysis rigorously establishes the stability and noise resistance of the model. To validate its effectiveness, the MNR-LCCZNN is applied to numerical simulations involving multiple constraints, as well as trajectory control experiments on the UR3e robotic arm. These experiments are conducted under both non-harmonic and harmonic noise interference. The results demonstrate that the proposed model delivers superior accuracy, robustness, and practical applicability across a range of representative noise scenarios.

Deep neural networks (DNNs) are highly susceptible to adversarial examples. By adding imperceptible perturbations to benign images, adversarial examples can mislead models into producing incorrect outputs, thereby posing significant security risks to DNN-based applications. However, most existing adversarial attack methods still achieve limited attack success rates. Focusing on black-box transfer attacks, we propose two novel approaches. First, TNI-FGSM aggregates gradients from the forward, backward, and current directions, enabling more stable updates and yielding more optimal perturbation directions. Second, NRSM initially performs random sampling on the input, followed by neighborhood resampling on both sides of the previous iteration's gradient. This process captures richer information, facilitates the discovery of optimal local extrema, and enhances transferability. Experiments conducted on ImageNet across conventional CNNs, four types of vision Transformers, and robust models demonstrate that NRSM consistently outperforms baseline methods. When Inception-v3 (Inc-v3) is used as the local model, NRSM achieves attack success rates (ASRs) that are 11.9% and 15.4% higher than LETM on Swin and HGD, respectively. Under the ensemble setting, NRSM attains an average ASR of 96.5%, surpassing Admix by 6.1%. Code is available at https://github.com/BreenoWH/NRSM-TNI.

One of the prevalent types of malignancy that affects women is called breast cancer. Early identification and treatment of cancer lessen the mortality rate. It is more common in women than men, which significantly increases the death rate of women. Prognosis and early identification are essential for ultimately increasing survival rates. The developed model is used to create effective techniques for detecting cancer in the women's at prior stage utilizing mammogram scans. The breast X-Ray images is also called as mammogram image. To recognize breast cancer from Mammogram images, an intelligent cancer framework is implemented using deep learning. The images needed for the classification process is collected from standard datasets. Next, the gathered images are provided to the segmented phase. Here, the segmentation is performed using a Weighted Adaptive Mask Region-based Convolutional Neural Network (WAM-RCNN), where the weights are optimally tuned using the Enhanced Fennec Fox Optimization (EFFO) algorithm. The segmented outcomes are fed to Vision Transformer-based MobileNet with Long Short Term Memory (ViT-MobLSTM) for the “detection and classification of breast cancer.” Finally, the suggested model is correlated with numerous metrics to find the competence of the model. The validation result shows that the developed model outperforms with effective outcomes.

This paper focuses on data-driven fault detection for multi-agent systems subject to actuator faults. For fault detection, a quantized data-driven fault detection scheme is constructed by making full use of data quantization information, which can reduce the computational burden and overcome dependence on a precise system mathematical model. Moreover, the developed method has the fault detection capability, where each agent can detect faults in all other agents within the topology network. A simulation study is conducted in the final section to validate the effectiveness of the proposed scheme.

Zeroing neural networks are widely applied in engineering fields. However, in practical scenarios, the problems they face often exhibit characteristics of uncertainty and fuzziness, and zeroing neural networks have obvious deficiencies in handling such uncertain problems. To address these deficiencies, researchers have combined zeroing neural networks with fuzzy systems. By leveraging the inherent advantages of fuzzy systems in dealing with uncertainty and fuzziness, this integration provides an effective way to expand the application scenarios of zeroing neural networks. This paper reviews the fusion and practical applications of two types of fuzzy systems (namely the Mamdani fuzzy system and the Takagi-Sugeno fuzzy system) with zeroing neural networks, aiming to offer references for subsequent research in this direction.

Autonomous driving technology is rapidly advancing and expected to revolutionize the transportation industry shortly. Although much research on trajectory tracking tasks has been carried out, it is essential to study how to accomplish assigned tasks under the disturbance of noise, especially for periodic noise, which is inevitably generated by electrical or motor interference in scenarios involving electronic circuits. For example, the suspension system constantly compresses and rebounds as the road surface changes during vehicle operation, generating the periodic noise. This paper establishes a kinematic model for the controlled vehicle and formulates the control problem as an optimization task using model predictive control (MPC). Subsequently, a periodic noise-suppressed neural network (PNSNN) model is proposed based on ordinary differential equations (ODEs) to achieve vehicle control. It is worth pointing out that the PNSNN model considers periodic noise from the perspective of harmonic expansion, thus effectively eliminating its interference. In addition, convergence analyses are conducted on the PNSNN model both with and without periodic noise. Finally, simulations and experiments validate the effectiveness and stability of the proposed PNSNN model.