Main problems of scanning electron microscopy which arise when working with low energies of electron probes are discussed. Ways of updating parameters of low-voltage electron microscopes through the development of mathematical simulation methods and software for optimizing electron-optical schemes of these microscopes are considered in detail.
Motor imagery brain-computer interfaces decode the user's intention from electroencephalography (EEG) signals. Though recent studies have shown the efficacy of a transformer architecture for this task, the standard self-attention in the transformer requires $O\left(N^{2} D\right)$ time complexity. Moreover, it is frequency-agnostic since it does not take into account the frequency structure of neural activities. This paper proposes Spectral-Separable Self-Attention (S3A), a novel frequency-aware attention mechanism. The S3A model decomposes token sequences into learnable spectral bands via differentiable Gaussian masks in the Fourier domain and applies separable selfattention for each frequency band and then fuses the results through learned gating with spectral residual bypass. Integrated into the Shallow Mirror Transformer architecture, SMT-S3A reduces the time and space complexity of self-attention from $O\left(N^{2} D\right)$ to $O(N \cdot D \cdot B)$ where $B \ll N$. Evaluation on BCI Competition IV Datasets 2a, 2b and PhysioNet, the SMT-S3A model achieves classification accuracy of 70.45%, 81.77% and 81.75%, respectively, under the leave-one-subject-out (LOSO) cross-validation setting. Notably, SMT-S3A model achieves this performance with as few as 32 K model parameters. Though the proposed S3A model is not given any frequency band information of neural activities in the training phase, the learned center frequency of the model converges to the frequency bands related to the well-known neural oscillatory components (e.g., mu, beta, and gamma band) of the corresponding motor-imagery tasks. This property provides posthoc interpretability of the proposed model. Also ablation studies show that both learnable spectral decomposition and band-wise self-attention contribute to the classification performance of the proposed SMT-S3A model.
The purpose of this article is to study the current state of Russia's energy security and identify ways to strengthen it. The article analyzes the most relevant indicators of Russia's energy security, which shows that the country's energy security is at an appropriate level. The article provides a detailed analysis of the electricity production per capita indicator and makes a forecast using the analytical smoothing method for the next two years. Based on the analysis, the article suggests ways to strengthen Russia's energy security, which will help protect the fuel and energy sector from internal and external threats.
Driving scene parsing is critical for autonomous vehicles to operate reliably in complex real-world traffic environments. To reduce the reliance on costly pixel-level annotations, synthetic datasets with automatically generated labels have become a popular alternative. However, models trained on synthetic data often perform poorly when applied to real-world scenes due to the synthetic-to-real domain gap. Despite the success of unsupervised domain adaptation in narrowing this gap, most existing methods mainly focus on global feature alignment while overlooking the semantic structure of the feature space. As a result, semantic relations among classes are insufficiently modeled, limiting the model’s ability to generalize. To address these challenges, this study introduces a novel unsupervised domain adaptation framework that explicitly regularizes semantic feature structures to significantly enhance driving scene parsing performance in real-world scenarios. Specifically, the proposed method enforces inter-class separability and intra-class compactness by leveraging class-specific prototypes, thereby enhancing the discriminability and structural coherence of feature clusters. An entropy-based noise filtering strategy improves the reliability of pseudo labels, while a pixel-level attention mechanism further refines feature alignment. Extensive experiments on representative benchmarks demonstrate that the proposed method consistently outperforms recent state-of-the-art methods. These results underscore the importance of preserving semantic structure for robust synthetic-to-real adaptation in driving scene parsing tasks.
The channel capacity of a MIMO communication system in a lossy inhomogeneous environment is investigated. The communication utilizes a microwave photonic signal generation principle, which first generates a radio signal and then modulate the optical radiation intensity with the radio signal. A model for the propagation of incoherent optical radiation in an inhomogeneous medium with losses has been developed, taking into account the attenuation of optical radiation and the refractive index of the inhomogeneous medium. The influence of the parameters of a heterogeneous medium with losses on the capacity of a MIMO communication channel has been studied. The study provides an original formulation of the propagation model that consistently incorporates both loss mechanisms and refractive index inhomogeneity. Obtained results demonstrate the sensitivity of channel capacity to key medium parameters, thereby establishing the novelty and practical relevance of the proposed approach.
