English and Foreign Languages University

The research introduces a secure system using privacy-preserving multi-modal biometric cryptography to confront major security flaws in biometric authentication over the cloud. The system uses both fingerprint and facial recognition along with strong cryptography to guard biometric data as it is sent and kept. Spreading, repeating and tampering with the system using false identities are mitigated by liveness checks, encrypting templates, using valid authentication keys during each use and through asymmetrical digital signatures. Simulations of over 1,000 authentications found that the FAR fell from 4.2% to 0.6% and the FRR reduced from 5.8% to 1.2%. In addition, the security system dealt with 95% of fake attempts and correctly agreed with genuine users in 99.1% of cases. Having encryption and authentication methods in place greatly cut the likelihood of man-in-the-middle attacks by 98%. The results suggest that the proposed system enhances security, provides privacy and maintains excellent user authentication rates in cloud biometric systems.

Future wireless systems require multiple access strategies capable of handling vast numbers of users and delivering fast data rates simultaneously. Code-domain NOMA attracts attention owing to its capacity to allocate resources and distinguish users by varying the interleaving order rather than frequency or time division. We present an enhanced interleaver structure formulated to boost the capacity and reliability of code-domain NOMA systems in future 6G wireless networks. We combine an adaptive interleaving module with a powerefficient transmission scheme to minimize inter-user interference and boost the performance of the multi-user receiver. The proposed approach considers QoS requirements and uses multi-objective genetic algorithms to address the resource allocation problem when faced with non-convex optimization objectives. The simulations showed higher bit error rates, greater energy savings and improved performance across a range of channel scenarios. The paper emphasizes that optimal interleaver design plays a critical role in achieving superior efficiency and device scaling in code-domain NOMA. These results drive further insight into how individualized encoding approaches shape the performance of emerging IoTintegrated networks in the context of 6 G deployment. According to the design, the system provided an 85.7 % maximum throughput efficiency and recorded a BER of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{5 1. 2 \%}$</tex> for AWGN. As a result, this system beats traditional interleavers in providing secure and easy-to-understand data transmission.

The work introduces a flexible design that ties together advanced methods for querying and deep learning with Apache Cassandra, for better management of large data sets. By using Cassandra's decentralized approach, the system can handle many different types of data without centralizing all the information. Researchers have introduced a hybrid system that pulls concepts from NoSQL to enhance the way users query and utilize data. Deep learning is also used to understand how queries are made and quickly use and allocate resources so the service remains responsive and error-free. The system is shown through testing to achieve high performance and low delay regardless of the workload, and is more advanced than traditional SQL approaches when it comes to scaling and recovering from failures. By mixing AI-driven optimization in the data management process, it becomes possible to add smart, automatic management to big data in many places. As a result, further advancements in data systems can be built using this approach as a base.

Centralized access control architectures demonstrate serious scalability problems and vulnerability to single points of failure along with privacy risks during IoT network expansion. The recently developed Blockchain and Federated Learning based Access Control Technology (BFL-ACT) serves as a distributed access control solution for IoT networks that uses blockchain technology combined with federated learning and cryptographic methods to enhance participant privacy protection and manage vast network sizes effectively. The protection of sensitive information and model updates during both training and inference phases utilizes integrated homomorphic encryption alongside lightweight cryptographic primitives. This paper demonstrates the system design and protocol flows for BFL-ACT and develops its threat model before constructing a prototype IoT testbed for performance benchmarking. The BFL-ACT approach demonstrates superior intrusion detection capabilities and increased robustness against attacks by malicious actors compared to standard centralized and non-blockchain federated learning systems while maintaining low communication costs and near real-time inference speeds ideal for practical IoT systems.

The World Health Organization (WHO) identified Gliomas as the most widespread brain tumour category. Toxicity levels combined with growth rates serve as the basis for dividing glioma malignancies into four distinct categories. The accurate grading of tumors determines which treatment method could be best suited including chemotherapy, surgery or radiation therapy. The use of magnetic resonance imaging (MRI) images to separate and grade tumours works well yet demands significant time and money and includes intricate management variations. The objective of this study is to meet the need for precise and reliable tumour classification by proposing a computational method for glioma grading that leverages an ensemble of Convolutional Neural Networks (CNNs). The proposed method introduces two ensemble models named Mean-Ensemble-CNNs & NN-Ensemble-CNNs that merge the pretrained CNN models ResNet-101, ResNeXt-50, Inception-v3 and DenseNet-161. Neural Network Ensembles combine deep features from multiple algorithms through a neural network to identify gliomas while Mean Ensembles calculate the average predictions from various algorithms to detect gliomas. The following sections provide a detailed analysis of Mean-Ensemble-CNNs. The system utilizes advanced data preparation and augmentation methods to fully exploit available informational resources. The technique reduces dependence on invasive biopsies while enabling flexible clinical treatment planning. Automatic segmentation and grading of tumours makes this process possible. The proposed NN-Ensemble CNNs achieve the top score with 96.21% accuracy which surpasses all individual models.

Through the provision of individualized material recommendations, recommendation systems are an essential component of online education. In this paper, hybrid optimizing techniques are proposed with the goal of improving the accuracy of recommendations. These techniques involve mixing the Imperialist Competitive Algorithm (ICA) with Simulated Annealing (SA) and Particle Swarm Optimization (PSO). In the ICA-SA approach, exploring is improved through the use of ICA, while SA is utilized for the purpose of local refining. Within the framework of ICA-PSO, autonomous nations that employ PSO-based movement techniques in order to improve their flexibility are presented. The application of these strategies improves the performance of deep neural networks while performing recommendations jobs. In a number of different test instances, the results show that ICA-PSO-DNN works better than other approaches that are currently in use. It achieves greater accuracy, recall, and F-measure. The ICA-PSO-DNN algorithm displays improved recommendations accuracy in comparison to both the logistic regression and solo deep learning systems currently available. A robust selecting features and better efficiency are both guaranteed by the hybrid technique, which efficiently strikes a balance between studying and exploiting respectively. The research sheds light on the influence that bio-inspired optimized methods have on the improvement of e-learning recommendation systems and offers beneficial insights into how these methods might be applied in contexts with vast amounts of data. Through the demonstration of the efficacy of hybrid optimization methodologies in enhancing recommended accuracy, this study makes a contribution to the continuous advancement of sophisticated system suggestions in the field of digital learning. Compared to existing models, the suggested technique yields a remarkable 98% accuracy.

This work investigates a unique technique to strengthen neural networks against misleading adversarial attacks, involving masking features. A major problem with Adversarial attacks is that they quietly switch parts of the data to mislead the AI system at inference. The method proposes to create exploitations by using K-Means Clustering and Class Activation Mapping (CAM). A random masking procedure is used during training to obfuscate the network of neurons with trivial characteristics, preventing reliance on them in decisionmaking. It is chosen for its ease of use, previously set performance goals and speed at which experiments can be done. A basic unmasked model is established in the study, followed by experiments using different masking rates. Standard attacks such as FGSM and PGD are used for detailed evaluations, checking how accuracy, precision, recall and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{F 1}$</tex> score are affected. From the experiments we see that correct application of feature masking greatly improves resistance to adversarial attacks, without much loss in model performance. The graphs of crucial metrics and images taken across the eras demonstrate the success of this approach in ensuring the model security and prediction efficiency are balanced. The suggested approach (CAM + K-Means + Masking) has a total accuracy of 94.5%. This increased accuracy demonstrates the model's robust ability to accurately categorize both regular and hostile inputs. The calculated equilibrium among accuracy, recall, and minimal error rates enhances this strong efficiency.

This paper examines the dialectics of “paracolonialism” and “wet globalization”—terms that have close affinity to blue humanities and ocean literature—through the east-west encounter depicted in the novel Ghost of Malabar (2022) written by Soumya Ayer. The ghosts in the novel are projections of tyranny of the colonial past who continue to haunt even after Independence. “Paracolonialism,” a term introduced by Dilip Menon, argues for a paradigm that moves outside the nation-state and the linear progression of pre-colonial, colonial, and post-colonial periods. Instead, he focuses on epistemologies of the Global South developed through oceanic connections, highlighting how these shared experiences shape modernity (Menon, 2020, p. 2). “Wet globalization,” on the other hand, is a term coined by Steve Mentz that offers a reminder that globalisation during and after the fifteenth century has “always contained the ocean as its material substrate and mechanism” (Mentz, 2015, p. xxix). Mentz argues that understanding globalization requires an oceanic perspective, recognizing the historical and ongoing impacts of maritime exchanges on global ecological systems. Thus, although both terms subvert the hierarchy of land by thinking of ocean, sea, river, and waterways—a critical discourse under the domain of blue humanities—that have contributed to the production of globalized cultures, it is not free from dichotomy. In this paper, I argue that Menon’s framework offers more innovative perspectives on modernity by questioning both elite and subaltern perspectives that coexisted with colonial modernity, enabling a recovery of identity from the harm caused by colonialism. While Menon used the term paracolonialism to come out of the tragic triad, Mentz emphasizes that we were never out of this tragic triad, and our current globalization still thrives on the ocean that has its roots in the precolonial era to this day.

This study examines Angela Carter “The Bloody Chamber” in the light of Martha Nussbaum’s theory of Objectification. The analysis of the short story discloses the devastating impacts of objectification on the lives of women characters by transforming their existence into mere objects of male pleasure and desire. This research explores how women are depicted as commodities, possessions, and passive objects and how these portrayals of women reinforce cultural attitudes towards female sexuality. By investigating the multifaceted nature of objectification in the selected literary text, this research work offers a comprehensive understanding of the methods in which objectification operates in the literary text and societal context. The findings of this research work reflect that socially constructed roles for sexes empower men to objectify women for their needs and purposes.

The retina acts as the main entrance point that allows visual data to enter the brain. Diabetic retinopathy is a circumstance affecting the blood vessels and nerve tissues in the retina. In recent times DR is becoming a major problem of vision loss, especially among diabetic people. It is important to detect it early and well-timed intervention can help prevent further vision loss. Retina Scan Disease Detection focusses to ensure a promising tool for DR prediction and aims to create a model that detects the existence of Diabetic Retinopathy and not only classifying them into mild, moderate, severe and proliferative depending on the severity level but also providing the output in a descriptive manner describing the conditions of fundus image. This paper employs a mixture of Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs). The hybrid CNN-RNN model contains benefits of transfer learning as well as Long Short-Term Memory (LSTM). The Diabetic Retinopathy is detected by using the CNN and LSTM is used for generating a descriptive sentence based on the condition of the image. The objective is to extend the state of the art of CNN-LSTM for three-dimensional dataset formation for the accurate prediction of diabetic retinopathy from the retinal images. The model is evaluated using BLEU scores The BLEU-1 score is nearly 0.86. Comparative analysis against existing approaches demonstrates superior performance in accuracy as well as generating sentences based on the severity of the disease. The model is able to generate a descriptive sentence about the condition of the retinal fundus image based on any severity level. The model has been evaluated on BLEU scores and has a BLEU score accuracy of about 86.17%.The outcomes prove that the automatic system not only reaches expert-level accuracy but also improves diagnostic efficiency, presenting a feasible solution for large-scale retinal screening in clinics and remote areas.