Modern computer networks require robust protection against increasingly sophisticated cyber threats. Intrusion Detection Systems (IDS) play a critical role in safeguarding the network perimeter and system boundaries by identifying malicious activities and unauthorized access. Machine learning and deep learning techniques have significantly improved intrusion detection performance; however, their effectiveness largely depends on appropriate model optimization and parameter selection. Although machine learning and deep learning have significantly advanced intrusion detection systems, their performance remains highly dependent on hyperparameter selection and optimization strategies. Conventional optimization methods often suffer from high computational cost and limited search capability, motivating the adoption of metaheuristic optimization algorithms to improve both detection accuracy and model generalization. This paper proposes two complementary optimization-based frameworks for binary intrusion detection using the UNSW-NB15 benchmark dataset. The first framework integrates Particle Swarm Optimization (PSO) with stacking ensemble learning to optimize the hyperparameters of three ensemble configurations, namely ET-RF, ET-DT, and RF-DT, constructed from Extra Trees (ET), Random Forest (RF), and Decision Tree (DT) classifiers. The second framework employs the Artificial Bee Colony (ABC) algorithm to directly optimize the weights and biases of a Dense neural network without relying on the conventional backpropagation algorithm. Experimental results demonstrate that the PSO-optimized RF-DT stacking model achieved the highest classification accuracy of 99.53
Anomaly detection in water distribution networks (WDNs) using unsupervised methods faces a fundamental limitation: the inability of the classic Isolation Forest (IF) algorithm to locate the semantic origin of detected anomalies and quantify their severity in an actionable manner within environments whose operational nature is constantly evolving. This paper proposes Contextual Isolation Forest (CIF), an adaptive extension of IF designed to operate on dynamic systems with continuous sensor data streams. CIF integrates three sequential procedures: domain-specific formal data modelling, empirical estimation of hyperparameters via controlled contamination, and contextual explainability through semantic partitioning based on domain expert criteria, functional properties of the system and geographical distribution. The method introduces two new anomaly indicators, Recurrence (RI) and Intensity (II), which extend the output of the classic IF from < score > to < score, RI, II>, enabling prioritisation and diagnosis without additional post-hoc analysis, and laying the foundations for self-adaptive monitoring strategies. The model was validated on 33,096 real-world records, derived from 11 sensor variables distributed across 4 geographical zones of a WDN in south-eastern Spain. The results demonstrate the stabilisation of detection metrics with just 10 trees and a threshold of 0.70, identifying 5 critical anomalies (RI ∈ [0.38, 0.63] and II ∈ [0.00, 0.77]) with explicit localisation by zone and type of measurement. CIF complements IF, EIF and DIFFI by adding operational explainability through semantic subspaces, without requiring labels or assuming prior distributions, constituting a viable component for anomaly detection systems in continuously evolving industrial environments.
The large-scale deployment of Generative Artificial Intelligence (AI), predominantly propelled by Large Language Models (LLMs), and simultaneous rapid progress in fault-tolerant quantum computers have motivated research into integrating quantum algorithms with the Transformer architecture. We introduce QBET (Quantum-Based Enhanced Transformer), a novel architecture that combines trainable position-aware quantum mixing, sparse attention mechanisms, quantum-enhanced attention selectively applied to important tokens, and Mixture-of-Experts feed-forward networks. In experiments on the Penn Treebank (PTB), WikiText-2, and WikiText-103 datasets, QBET achieved perplexity (PPL) scores of 93.3, 211.5, and 145.8, respectively, converging within five epochs on PTB, six epochs on WikiText-2, and five epochs on WikiText-103, outperforming both classical and quantum baseline models. All experiments are conducted via classical simulation, consistent with standard practice in quantum machine learning research. This work contributes an empirical foundation for hybrid quantum-classical language modeling, demonstrating that quantum-inspired architectural components can achieve competitive performance with classical baselines at small scale. We position QBET as a hybrid-inspired architecture whose performance gains are partially attributable to quantum circuit structure, with stronger conclusions deferred to future work at larger scale and on real quantum hardware.
DNA-based approaches, particularly COI barcoding and genomic workflows, have contributed to establishing integrative taxonomy as the current standard, but have also been promoted as the primary, and sometimes sole, avenue for species discovery and delimitation in hyperdiverse insect groups such as Hymenoptera. Here, we examine the recent emphasis on DNA-based taxonomy and its implications for species diagnoses and long-term nomenclatural stability. We argue that molecular data are indispensable, but that integrative approaches combining morphological evidence with mitochondrial and nuclear markers provide the most robust framework for efficient species discovery and stable, testable species hypotheses. From our shared perspective of studying megadiverse parasitoid wasps, we further argue that the success of integrative taxonomy ultimately depends on researchers with both methodological skills and deep knowledge of the diversity, biology and evolution of their focal taxa. The global decline in expert taxonomists therefore represents a major challenge for biodiversity research, making the training of a new generation of integrative taxonomists essential.
From a collection of IsopodaAnthuroidea the new species Eisothistos minahasaensis sp. nov ., Mesanthura ratulangii sp. nov ., Mesanthura heliolatra sp. nov ., Haliophasma annaclericiae sp. nov ., Apanthuretta markuslasuti sp. nov ., Apanthuretta pinchans sp. nov ., Amakusanthura mamuajae sp. nov ., and Apanthurella cuspidobtusa gen. nov. et sp. nov . are described. The species were found in near-shore samples of coral rubble and sediment in North Sulawesi (Indonesia). The genus Haliophasma Haswell, 1881 is compared with Malacanthura Barnard, 1925 and the genus Exanthura Barnard, 1914 is reinstalled to accommodate species with a basal hook on the A1. Nemanthura Wägele, 1981 is synonymized with Exanthura Barnard, 1914. A new classification of the Apanthura group, including the diagnosis of the new genus Apanthurella gen. nov ., is provided together with lists of hitherto described species. Diagnoses for the genera Haliophasma , Malacanthura , Exanthura , Apanthura , Apanthuretta , Apanthuropsis , Amakusanthura , and Apanthurella are presented.
