The rising frequency and sophistication of cyberattacks underscore the urgent need for adaptive Network Intrusion Detection Systems (NIDS). Traditional systems, though effective against known threats, struggle to adapt to evolving traffic patterns and novel attack vectors. A central challenge is catastrophic forgetting, whereby models overwrite prior knowledge when updated with new data, undermining their reliability in dynamic environments. Continual learning (CL) offers a promising paradigm for mitigating this challenge by enabling incremental adaptation without complete retraining. Yet, its systematic application and evaluation within NIDS remain limited. To address this gap, we introduce the Anomaly-Aware Class-Incremental Learning (ACIL), a framework designed to evaluate CL strategies under realistic network conditions. ACIL operationalizes anomaly-awareness by structuring task streams to reflect natural traffic imbalance: benign flows dominate early stages, while rare attack classes are introduced incrementally. This setup more faithfully represents the challenges of evolving intrusion detection by reflecting natural traffic imbalance, in contrast to uniform or artificially balanced learning. We assess ACIL on three widely recognized datasets CIC-IDS-2017, CIC-IDS-2018, and NSL-KDD using multiple state-of-the-art CL methods. Results show that CL strategies enhance adaptability and mitigate forgetting relative to static baselines, but notable trade-offs persist, especially for minority and late-emerging classes. These findings establish ACIL as a standardized evaluation framework for continual learning in cybersecurity and emphasize the need for anomaly-aware mechanisms to develop resilient NIDS in dynamic threat landscapes.

Modern network environments evolve rapidly, exposing intrusion detection systems (IDS) to continually emerging and previously unseen cyberattacks. Most existing IDS models cannot incorporate new attacks without full retraining, and they typically suffer from catastrophic forgetting, limited generalisation, and poor detection of zero-day or rare threats. We propose CLEAR (Continual Learning with Embedded Attack Replay), a lightweight framework that integrates continual learning, semantic representation, and hybrid detection. CLEAR learns new attack classes incrementally while preserving prior knowledge through an efficient replay strategy. A contrastively aligned language model produces semantic embeddings of network flows, enabling generalisation to unseen attacks, and a drift-aware module adjusts detection thresholds as traffic patterns evolve. The CLEAR unifies semantic-guided replay, supervised-unsupervised hybrid detection, and drift-sensitive adaptation within a single continual learning architecture. This design supports stable performance under limited data, shifting traffic distributions, and strict zero-day conditions. Experiments on four benchmark datasets (CIC-IDS-2017, CIC-IDS-2018, BoT-IoT 2018, and IoT-23) show that CLEAR achieves strong overall performance (F1 ≈ 0.89, average FPR ≈ 0.08), reduces catastrophic forgetting by up to 24%, and attains zero-day detection rates of 88%. CLEAR also performs well on rare attacks (F1 ≈ 0.83) and few-shot classes (F1 ≈ 0.85), demonstrating its suitability for adaptive and real-time intrusion detection in dynamic network environments.

Network Intrusion Detection Systems (NIDS) are critical for cybersecurity, yet they consistently fail to detect novel, rare attacks due to the extreme scarcity of labeled data. While few-shot learning (FSL) offers a promising solution, existing methods often rely on single, static encoders that develop “blind spots”, failing to generalize across the diverse signatures of modern threats. To overcome this critical challenge, we propose a novel FSL framework built on a robust metric-learning paradigm. We have developed a novel Gated Ensemble Encoder, an architecture that learns a committee of “specialist” feature extractors managed by a dynamic gating mechanism. This design produces a dynamic embedding for each input by adaptively weighting the specialists’ contributions on a per-sample basis, thus eliminating the blind spots of monolithic models. To create a geometrically sound representation space, our architecture is meta-trained using a Prototypical Network strategy on a diverse set of common, high-volume attacks. During evaluation, the model demonstrates the ability to learn and classify entirely new, unseen attack classes from as few as five examples. Tested on the comprehensive NSL-KDD, CIC-IDS 2018, and CIC-ToN-IoT datasets, our framework achieves a Macro F1-Score of up to 0.84 in challenging 4-Way, 5-Shot scenarios and successfully detects attack classes that were previously undetectable. These results validate our dynamic embedding strategy as a highly effective and practical solution for zero-day threat identification in real-world, data-scarce environments.

Sequential recommendation has achieved remarkable progress with self-attentive architectures such as SASRec (Self-Attentive Sequential Recommendation), yet existing approaches often underutilize semantic relationships among items and fail to model temporal dynamics effectively. These limitations reduce the ability of existing systems to capture nuanced user preferences in real-world settings. To address these issues, this work introduces two novel frameworks: one for effective semantic integration and another for a semantic and temporal-aware hybrid embedding generation that enhances the representational capacity of SASRec. These frameworks construct a semantically rich hybrid matrix utilizing Markov State Transition probabilities, Cosine similarity, Personalized PageRank(PPR) based normalization and additionally Time-weighted decay information for the temporal variant. The constructed hybrid matrix is smoothed using a graph convolutional network (GCN) to generate item embeddings, which are then passed to the transformer-based sequential recommendation model SASRec for next-item prediction. Evaluated on three real-world benchmark datasets (MovieLens, Yelp and Amazon Beauty), our proposed temporal variant achieves up to 10.4

Network intrusion detection aims to identify anomalous activities in network traffic, while continual learning (CL) methods strive to preserve past knowledge and adapt to evolving threats. Memory replay-based CL approaches have been widely used and proven effective at mitigating catastrophic forgetting. However, previous research has primarily focused on addressing class imbalance and has largely relied on augmented and random memory replay strategies, which introduce significant computational overhead and limit practicality in real-time applications. To overcome these challenges, we propose Task-Aware Memory Replay (TAMR), a novel framework that prioritizes past experiences based on their relevance to the current task. By dynamically adjusting the importance of replayed samples, TAMR balances the integration of new attack patterns with the retention of critical historical knowledge, ensuring resilience against evolving threats and variations in normal traffic. Unlike traditional methods that employ random selection or augmented replays, TAMR selectively replays high-impact experiences, thereby optimizing memory usage and improving adaptability. Our experiments demonstrate that TAMR achieves real-time adaptability across five distinct NIDS datasets, ultimately delivering superior performance and computational efficiency in detecting even unknown attacks in dynamic network environments. In general, we highlight the potential of memory-based replay strategies for continual learning in detecting unknown attacks using a task-aware approach.

Ultrasonic velocity (U), density (rho) and viscosity (eta) have been measured for morpholine with ortho-xylene in isopropanol at 303K, 308K & 313K. From these experimental values, the adiabatic compressibility (beta), acoustic impedance (Z), free length (Lf), free volume (Vf), internal pressure (pi i), cohesive energy (CE), Lenard Jones Potential (LJP) and viscous relaxation time (r) have been calculated and their importance are discussed. Formation constant (K) values of the compounds have been determined by using a modified Bhat equation. The thermodynamical parameters have also been computed and the complexes are thermodynamically stable as noticed from their free energy of formation values. The theoretical values of ultrasonic velocity have been calculated at respective temperatures using the Nomoto's relation (UNOM), Impedance relation (UIMP), Ideal mixing relation (UIMR) and Junjie's relation (UJUN) and these values are further confirmed by Chi-Square test analysis. The molecular interaction parameter has been calculated by using the experimental and theoretical ultrasonic velocity values.

Traditional network intrusion detection systems (NIDS) face significant scalability challenges due to the vast amount of data generated by Internet of Things (IoT) devices, compounded by growing privacy concerns. Federated Learning (FL) has emerged as a promising solution, offering a distributed, privacy-preserving paradigm that enables Deep Learning (DL) models to be trained locally, thereby mitigating privacy risks associated with centralized data processing. However, conventional FL strategies come with inherent limitations. First, they require all clients to use the same model architecture, making personalized learning difficult particularly in Non-Independent and Identically Distributed (Non-IID) heterogeneous data settings. Second, the weight aggregation process in FL introduces significant communication overhead, potentially slowing down training. While encryption techniques such as homomorphic encryption and differential privacy enhance security, they also increase computational costs and can still reveal data distribution patterns if compromised. These challenges are further exacerbated in dynamic IoT environments, where evolving attack types continuously alter data distributions. To address these issues, we propose a Personalized Federated Hypernetworks-based aggregation strategy for Intrusion Detection Systems (PerFedHypID). Unlike conventional FL approaches that rely on weight-based aggregation, PerFedHypID utilizes embedding vectors, which are computationally lighter and enable enhanced personalization. Our method leverages personalized layers and hypernetwork-based aggregation to achieve both efficiency and adaptability. We extensively evaluate PerFedHypID on the CSE-CICIDS-2018 and UNSW-NB-15 datasets under various non-IID heterogeneous settings. The results demonstrate that our approach outperforms state-of-the-art personalized federated learning algorithms, offering robust performance and improved adaptability in dynamic IoT environments.

Medical imaging with its high resolution and detailed anatomical insights plays a crucial role in early cancer detection and diagnosis. However, achieving precise segmentation remains challenging, particularly with conventional deep learning models that demand substantial computational resources. To enhance the efficiency of cancer region segmentation, we propose a lightweight Generative Adversarial Network (GAN) model, based on a modified pix2pix architecture. The generator and discriminator in this architecture are optimized using group and spectral normalization techniques, improving gradient propagation while reducing model complexity and computational overhead. The proposed approach is evaluated on two distinct medical modalities: microultrasound for prostate and contrast-enhanced Magnetic Resonance Imaging (MRI) for hepatocellular carcinoma (HCC) segmentation. The performance of the proposed model is evaluated using the Dice coefficient and pixel accuracy, showing significantly higher values compared to conventional segmentation models for both prostate and HCC across two different imaging modalities. Furthermore, the proposed model achieves faster training times with improved efficiency, which is 17.8

Accurate network intrusion detection requires extracting relevant semantic features to minimize mis-classification and identify various kinds of attacks. Traditional models often struggle with uncertain traffic patterns, reducing their reliability. This work presents a Novel Semantic-Driven Meta-Learning Model - a hybrid framework that systematically refines categorization decisions. It uses advanced learning techniques to improve detection through a two-stage verification process. Our approach enhances intrusion detection by integrating an attention-based model for semantic feature extraction and the Simple Neural Attentive Meta-Learner (SNAIL) for rare attack class detection. In the first phase, machine learning classifiers perform macro-classification to distinguish between normal and attack traffic. To reduce mis-classification, an additional classifier verifies the predicted class label using a refined subset of features. In the second phase, the SNAIL model conducts micro-classification, further differentiating attack classes with high precision. This dual-phase strategy is particularly effective in addressing the challenge of detecting rare classes in highly imbalanced network intrusion datasets. Using the NSL-KDD, CSE-CIC-IDS 2018, and CIC-ToN-IoT datasets, we assess our model, showing its better performance in precisely detecting unusual attack types while preserving a low false alarm rate with semantic characteristics. The proposed model demonstrates a higher detection rate for most of the rare classes considered in our study compared to state-of-the-art methods.

Federated learning (FL) offers a promising approach for training machine learning models with minimal data sharing, enhancing privacy and performance. However, building effective FL-based network intrusion detection systems (NIDS) remains challenging due to the need for large, diverse training datasets. Identifying rare attack types with limited instances is a persistent obstacle, and their detection is critical in cybersecurity. This research introduces a novel FL framework to address these challenges. By incorporating adaptive, personalized layers at the client level, the model reduces false alarm rates for zero-day attack types and improves the detection of rare classes. The model also leverages Transfer Learning (TL) to identify zero-day attacks, where client-specific gradients are collected and used to update a global model on the server side after multiple rounds of exposure to new data. The proposed sustainable framework aims to disseminate knowledge about rare attack types across clients through a server-based global model within the FL ecosystem. This study achieves two main objectives: (i) improving the detection of rare attack classes and (ii) identifying zero-day attacks in a NIDS context. Evaluations on the CSE-CICIDS-2018, Edge IIoT, and UNSW-NB 15 datasets, which encompass diverse class distributions, demonstrate that the proposed approach outperforms existing models in detecting and handling rare and novel attack types. The proposed model achieves 98.90% accuracy on CICIDS 2018, 98.70% on UNSW-NB 15, and 97.92% on Edge-IIoT, surpassing the FL-TL-CNN model by 2.78%, 1.51%, and 2.03%, respectively. These results highlight the effectiveness, robustness, and adaptability of the proposed approach in enhancing intrusion detection across heterogeneous network environments.

Network Intrusion Detection Systems (NIDS) are essential for identifying and mitigating malicious activities in network environments. As cyber threats evolve in complexity, traditional NIDSs often struggle to detect sophisticated attacks effectively, especially those involving intricate temporal and spatial dependencies within network traffic. The ability to capture these dependencies is crucial for reducing false positive rates and improving detection accuracy. However, existing models face significant challenges, including handling varying sequence lengths and capturing long-range dependencies, which are essential for accurate anomaly detection. This paper proposes a hybrid model that combines Sequence to Sequence (Seq2Seq) architecture with Convolutional Long Short-Term Memory (ConvLSTM) units to address these challenges. This hybrid model handles spatial and temporal dependencies by incorporating convolutional layers within LSTM cells. This enables the model to leverage the spatial feature extraction capabilities of Convolutional Neural Networks (CNN) alongside the sequential learning strengths of LSTM networks. In addition, to enhance the interpretability of the model, the proposed architecture integrates Explainable Artificial Intelligence (XAI) through Local Interpretable Model-agnostic Explanations (LIME). This approach provides insights into the model’s decision-making process, highlighting the temporal and spatial features that influence predictions and improving transparency in detecting anomalies. Experimental evaluations on benchmark datasets, including CIC-IDS 2017, CIC ToN IoT, and UNSW-NB15, demonstrate that the proposed hybrid ConvLSTM-Seq2Seq model outperforms existing methods in reducing false positives and achieving higher accuracy. This model offers a promising solution for NIDSs by improving detection capabilities and better handling complex temporal and spatial relationships in network data.

Sequential recommendation systems are designed to predict user's subsequent actions and recommend relevant items, particularly within session-based interactions where user preferences can change rapidly. A significant challenge in these systems is dealing with short sequences of user interactions, which often lack sufficient data to capture user preferences, leading to imprecise recommendations. Although various research efforts have aimed to improve the handling of short sequences, the presence of uninteresting items within these sequences continues to hinder the accuracy of the recommendation system. To address this challenge, we propose a novel approach, Session-based Uninteresting Item Removal for Enhanced Recommendations (SURE), which integrates Short-Sequence Enhancement with the innovative removal of uninteresting items using association rule mining. Our method incorporates a backward prediction model that generates pseudo-historical items, enriching short sequences while preserving relevant information. The core novelty of our approach lies in the targeted removal of uninteresting items through association rule mining. By systematically identifying and filtering out items deemed uninteresting based on user preferences, we refine the sequence data to ensure that only relevant and engaging items remain. This refinement process enhances the overall quality and accuracy of the recommendations. Our approach significantly improves recommendation performance, achieving an average gain of 7.31% in mean reciprocal rank (MRR) and higher Normalised Discounted Cumulative Gain (nDCG) and accuracy scores across multiple real-world datasets compared to state-of-the-art techniques. These advancements highlight the model's ability to deliver more accurate, relevant, and interpretable recommendations, demonstrating its potential for real-world applications.

Session-based recommendation systems often struggle to personalise recommendations because they focus on item sequences within a user's current session, overlooking valuable historical data and broader item relationships. While Heterogeneous Global Graph Neural Networks (HG-GNNs) can capture long-term and short-term user preferences, their complexity can be prohibitive. To address this, we propose a method that forms clusters of users based on their implicit preferences, reducing the complexity while still effectively modelling user preferences. By clustering users with similar implicit preferences, our method simplifies the recommendation process while maintaining the ability to capture diverse preferences. This clustering approach allows for more efficient learning within the GNN, balancing the need for accuracy with computational feasibility. Experiments on real-world datasets show that our approach maintains comparable accuracy to traditional HG-GNN methods while significantly reducing computational overhead. This balance of efficiency and effectiveness sets a new standard for session-based recommendation systems.

Federated learning has emerged as a powerful solution in network intrusion detection, proving highly effective in countering sophisticated attacks. This work delves into the intricate workings of a federated system, exploring its capabilities through a structured, three-stage workflow. The first stage involved data exploration, followed by model construction and testing, where critical parameters such as training epochs, data partitioning, and data proportions were adjusted to evaluate their impact on the model’s performance. In the final stage, feature exploration, an explainable AI framework was employed to assess feature contributions, with the premise that models utilizing the same dataset should assign similar importance to key features. A comparison between the centralized and federated models was conducted, revealing that both approaches aligned closely in their feature interpretations. Explainable AI brought enhanced transparency, ensuring that the model’s decision-making was consistent with practical, real-world reasoning. In conclusion, this study highlights the robustness of federated models in intrusion detection, offering valuable insights into their parameter optimization and confirming their potential as a reliable alternative to centralized systems. We have utilized CICIDS 2017 dataset to evaluate and demonstrate the results.

Federated learning (FL) is a potential method for training machine learning models that aims to minimize the sharing of data to maximize privacy and performance. However, constructing efficient FL-Network Intrusion Detection Systems(NIDS) requires a substantial quantity of varied training data. Identifying rare attack classes with few shots presents a growing obstacle in this domain, and their detection is a pressing concern. This research proposes a novel FL framework to address the problem by decreasing false alarm rates for unseen attack classes and enhancing the detection of rare classes with a few shots by adaptive personalized layers in FL at the client end. The proposed model employs transfer learning to detect zero-day attacks by gathering tailored client gradients and updating the global model on the server side after observing many rounds of new shots. This strategy aims to disseminate knowledge about rare attack classes to all clients by employing a global model on the server within the federated ecosystem. More precisely, this study has accomplished two significant feats: (i) showcasing the enhancement of identifying rare attack classes and (ii) detecting zero-day attacks in a Network Intrusion Detection System (NIDS) environment by conducting experiments in different scenarios. We thoroughly assessed our proposed approach using the CSE-CICIDS-2018 dataset across several class distribution scenarios. The results exhibit strong performance in identifying and managing novel and rare attack classes compared to current models.

Auto manufacturers and research groups are working on autonomous driving for long period and achieved significant progress. Autonomous vehicles (AV) are expected to transform road traffic reduction from current conditions, avoiding accidents and congestion. As the implementation of an autonomous vehicle ecosystem includes complex automotive technology, ethics, passenger behaviour, traffic management policies and liability etc., the maturity of AV solutions are still evolving. The proposed model to improve AV user experience, uses a hybrid AV Network of multiple connected autonomous vehicles which communicate with each other in an environment shared by human driven vehicles. The proposed Optimal AV Network (OAVN) solution provides better coordination and optimization of autonomous vehicles, improved Transportation efficiency, improved passenger comfort and safety, real-time dynamic adaption of traffic road conditions along with improved in-cabin assistance with inputs from various sensors. The true optimal solution for this problem, is to devise an automated guidance system for vehicles in an AV network, to reach destinations in best possible routes along with passenger comfort and safety. A custom informed search model is proposed along with other heuristic goals for better user experience. The results are analysed and compared to evaluate the effectiveness of the solution and identify gaps and future enhancements.

This paper investigates the effectiveness of attention mechanisms in speech classification tasks for predicting age, gender, and emotion. Despite the success of attention in multimodal systems, its applicability to speech context preservation remains underexplored. We examine this using the SEGAA model alongside our ASEGAA (Attention in Speech Emotion Gender and Age Analysis) models. Experiments on CREMAD and EMO-DB datasets reveal challenges in capturing speech context due to limited MFCC correlation. Speech attention models struggle to maintain context, leading to reduced accuracy compared to non-attention speech models. The results highlight the significance of feature selection and advocate for further exploration of alternative mechanisms. In terms of F1 scores, the multi-output SEGAA model achieves relatively high scores across age, emotion, and gender prediction tasks. However, attention mechanisms such as ASEGAA attention at end and ASEGAA Interspersed Attention exhibit fluctuating F1 scores, suggesting differing impacts on context preservation across the prediction tasks.

The maintenance, archiving and usage of the design drawings is cumbersome in physical form in different industries for longer period. It is hard to extract information by simple scanning of drawing sheets. Converting them to their digital formats such as Computer-Aided Design (CAD), with needed knowledge extraction can solve this problem. The conversion of these machine drawings to its digital form is a crucial challenge which requires advanced techniques. This research proposes an innovative methodology utilizing Deep Learning methods. The approach employs object detection model, such as Yolov7, Faster R-CNN, to detect physical drawing objects present in the images followed by, edge detection algorithms such as canny filter to extract and refine the identified lines from the drawing region and curve detection techniques to detect circle. Also ornaments (complex shapes) within the drawings are extracted. To ensure comprehensive conversion, an Optical Character Recognition (OCR) tool is integrated to identify and extract the text elements from the drawings. The extracted data which includes the lines, shapes and text is consolidated and stored in a structured comma separated values(.csv) file format. The accuracy and the efficiency of conversion is evaluated. Through this, conversion can be automated to help organizations enhance their productivity, facilitate seamless collaborations and preserve valuable design information in a digital format easily accessible. Overall, this study contributes to the advancement of CAD conversions, providing accurate results from the translating process. Future research can focus on handling diverse drawing types, enhanced accuracy in shape and line detection and extraction.

The parameters like ultrasonic velocity, density and viscosity of ternary liquid mixtures containing di-substituted benzene [o-methoxynitrobenzene (o-MNB)] with 1-alkanols (1-propanol/1-butanol/1-pentanol) in n-hexane medium have been measured experimentally at 303K for the equimolar concentration of 1:1 ratio and the concentration ranges from 1x10(-3) M to 1x10(-2)M. The normal acoustical properties and their excess properties such as excess ultrasonic velocity (U-E), excess acoustic compressibility (kappa(E)), excess free length (L-f(E)), excess free volume (V-f(E)) and excess internal pressure (pi(E)(i)) have been computed. The molecular interactions and their strength have been quantitatively assessed and the cyclic structure is proposed from the determination of thermodynamic parameters like formation constant (K), free energy of formation (Delta G(F)), enthalpy change (Delta H) and entropy change (Delta S) through the formation of the hydrogen bonded complexes. The results are discussed in terms of intermolecular interaction based on existence of carbon profile between the components present in the ternary liquid mixtures.

The objective measurements of the real-world optimization problems are mostly subject to noise which occurs due to several reasons like human measurement or environmental factors. The performance of the optimization algorithm gets affected if the effect of noise is higher than the negligible limit. The previous noise handling optimization algorithms use a large population size or multiple sampling at same region which increases the total count of function evaluations, and few methods work for a particular problem type. To address the above challenges, a Differential Evolution based Noise handling Optimization algorithm (NDE) to solve and optimize noisy bi-objective optimization problems is proposed. NDE is a Differential Evolution (DE) based optimization algorithm where the strategies for trial vector generation and the control parameters of DE algorithm are self-adapted using fuzzy inference system to improve the population diversity along the evolution process. In NDE, explicit averaging based method for denoising is used when the noise level is higher than negligible limit. Extending noise handling method enhances the performance of the optimization algorithm in solving real world optimization problems. To improve the convergence characteristics of the proposed algorithm, a restricted local search procedure is proposed. The performance of NDE algorithm is experimented using DTLZ and WFG problems, which are benchmark bi-objective optimization problems. The obtained results are compared with other SOTA algorithm using modified Inverted Generational Distance and Hypervolume performance metrics, from which it is confirmed that the proposed NDE algorithm is better in solving noisy bi-objective problems when compared to the other methods. To further strengthen the claim, statistical tests are conducted using the Wilcoxon and Friedman rank tests, and the proposed NDE algorithm shows significance over the other algorithms rejecting the null hypothesis.