
The growing prevalence of online abuse has increased the need for automatic cyberbullying and hate speech detection, particularly in under-resourced and morphologically complex languages with limited annotated data and language resources. This study presents a multilingual English–Lithuanian framework combining machine learning, deep learning, transformer models, three data augmentation techniques, and explainable AI to address class imbalance, limited data, and model transparency. Overall, transformer-based approaches outperformed traditional methods. For the English dataset, averaged results over five runs with different random seeds showed model-dependent performance. Syn-Aug achieved the best F1-scores for LR (0.8791 ± 0.0029), SVM (0.8810 ± 0.0028), and RoBERTa (0.9287 ± 0.0023), while contextual BERT augmentation achieved the best F1-scores for CNN (0.9124 ± 0.0020) and BERT (0.9274 ± 0.0021). The highest English performance was obtained with the soft-voting BERT-RoBERTa ensemble, with an F1-score of 0.9356 ± 0.0035. For Lithuanian hate speech classification, hold-out validation and five-fold cross-validation were employed and evaluated on the same independent test set. For the five-fold cross-validation, XLM-RoBERTa and Syn-Aug achieved the best accuracy (0.9458) and F1-score (0.9363), followed by XLM-RoBERTa with XLM-Aug (accuracy 0.9072; F1-score 0.9336). Under the hold-out protocol, XLM-RoBERTa with XLM-Aug achieved the best accuracy (0.824) and F1-score (0.8109). These results show that Lithuanian classification performance varied with augmentation and evaluation protocol. Cross-language transfer using semantically aligned binary labels further highlighted the importance of language-specific data. Overall, the proposed approach supports efficient, interpretable, and reliable multilingual cyberbullying detection.
Nowadays, synthetic aperture radars are widely used on various satellite imaging platforms. The high availability of SAR imagery makes it very popular in various automatic object detection systems. However, the visual interpretation of SAR data is difficult even for experts, which poses the primary challenge in preparing the data required for training and evaluating deep learning models. Therefore, using data augmentation techniques to expand the original training dataset is a popular approach. In this paper, a method of synthetic SAR object image generation is proposed using Principal Component Analysis. The presented technique is based on principal component perturbations and can be used for data augmentation to train object detection models. The numerical experiments performed with the YOLO model showed that the proposed method improved AP50 on the test set for transport aircraft detection, with statistical significance confirmed by bootstrap analysis.
The push–pull inverter topology inherently incorporates a transformer, making it suitable for electrosurgical generators without requiring an additional transformer. However, conventional push–pull topologies exhibit limitations, including high switch stress and the absence of a dedicated path for releasing transformer leakage-inductance energy, which may increase switching losses at high frequencies. This paper presents a bipolar pulse generator based on a push–pull inverter topology, offering a new engineering approach for electrosurgical applications. The prototype achieves soft-switching operation over a load range of 30–1000 Ω, reducing design complexity, extending load adaptability, and improving system reliability while meeting the stringent pulse-width control requirements of bipolar electrosurgical applications. In addition, based on an equivalent-impedance model of tissue welding, the operating modes and soft-switching mechanisms of the push–pull inverter under different dynamic load conditions are analyzed, and strict constraints relating to dead time, load resistance, and circuit parameters are derived. A bipolar pulse-generator prototype operating at 400 kHz was developed to validate the theoretical analysis. Experimental results show a conversion efficiency of 86–94% over a dynamic load range of 25–250 Ω. At a moderate nominal load, the system maintains an efficiency above 93% over an output-power range of 4.92–183.69 W, achieving high-efficiency operation across a power range relevant to electrosurgical applications. While maintaining tissue-welding quality, the proposed system improves overall efficiency, demonstrating the feasibility and engineering value of the topology and control strategy for bipolar electrosurgical generators.
In power grids, the introduction of renewable energy may affect frequency stability. Therefore, this paper considers adjustable predefined-time sliding-mode frequency control of microgrids with hybrid energy storage systems (HESSs) and cyberattacks. First, by analyzing dynamics of the ultracapacitors, batteries and cyberattacks, models of the HESS and aperiodic DoS attacks are established. To compensate the disturbances induced by renewable energy, an adjustable predefined-time observer (APTO) is designed, which combines adaptive laws such that the observation error converges to the specified boundary with a predefined time that is regulated by a flexible scalar. Based on the observer, an integral sliding surface is designed. Then, by introducing a switched system, the stability of the sliding-mode dynamics is analyzed. Moreover, to guarantee reachability of the sliding surface, a predefined-time sliding-mode controller is designed. Compared with the existing results, the proposed controller is independent of the initial states. At last, the two-area microgrids are used to illustrate the effectiveness of the designed controller. Compared with SMC and FTSMC, the frequency deviation is reduced by 6.79% and 4.78% and the convergence time is reduced by 31.6% and 13.3%. It is shown that the proposed controller achieves fast convergence and strong robustness.
Whether edge-scale language models can call tools, and whether teams help, is understudied. We present a fine-grained diagnostic evaluation of tool calling on open-source edge models (0.6B–9B). The single-agent framework measures four dimensions and attributes every run to one of seven failure modes; parameter filling, not tool selection, separates single-agent models. The evaluation extends to multi-agent systems: an orchestrator-workers workflow over 30 MCP tools, a two-layer failure taxonomy of role delegation and worker execution, and a noise-floor protocol of 30 repeats with paired t-tests, exact permutation tests, and bootstrap intervals. Delegation rises steeply with orchestrator scale. Even a Qwen3-0.6B coordinator can delegate: a template-controlled sweep traces its zero delegation to the chat template attached to the model tag rather than to weight precision, and the same weights reach 12.5–22.1% under the template embedded in the model file. Orchestrator-workers teams do not significantly outperform a matched single model or best-of-3, an oracle-aided upper bound rather than a deployable selector, at any scale we tested: delegation overhead consumes the accuracy workers achieve. These findings hold on 4090 GPUs and reproduce on an Ascend 310P NPU. Cross-family runs reproduce the general directional pattern, and the tested orchestrator-workers teams provide no demonstrated advantage over the corresponding single-model baselines under this benchmark and deployment setup.