Doklady Mathematics最新文献

This paper presents the development and evaluation of methods for detecting cyberattacks on industrial systems using neural network approaches. The focus is on the task of detecting anomalies in multivariate time series, where the diversity and complexity of potential attack scenarios require the use of advanced models. To address these challenges, a transformer-based autoencoder architecture was used, which was further enhanced by transitioning to a variational autoencoder (VAE) and integrating normalizing flows. These modifications allowed the model to better capture the data distribution, enabling effective anomaly detection, including those not present in the training set. As a result, high performance was achieved, with an F1 score of 0.93 and a ROC-AUC of 0.87. The results underscore the effectiveness of the proposed methodology and provide valuable contributions to the field of anomaly detection and cybersecurity in industrial systems.

Handwritten character recognition (HCR) is a challenging problem for machine learning researchers. Unlike printed text data, handwritten character datasets have more variation due to human-introduced bias. With numerous unique character classes present, some data, such as Logographic Scripts or Sino-Korean character sequences, bring new complications to the HCR problem. The classification task on such datasets requires the model to learn high-complexity details of the images that share similar features. With recent advances in computational resource availability and further computer vision theory development, some research teams have effectively addressed the arising challenges. Although known for achieving high accuracy while keeping the number of parameters small, many common approaches are still not generalizable and use dataset-specific solutions to achieve better results. Due to complex structure, existing methods frequently prevent the solutions from gaining popularity. This paper proposes a highly scalable approach for detailed character image classification by introducing the model architecture, data preprocessing steps, and testing design instructions. We also perform experiments to compare the performance of our method with that of existing ones to show the improvements achieved.

In this paper, we present a novel methodology we call MDS-ViTNet (Multi Decoder Saliency by Vision Transformer Network) for enhancing visual saliency prediction or eye-tracking. This approach holds significant potential for diverse fields, including marketing, medicine, robotics, and retail. We propose a network architecture that leverages the Vision Transformer, moving beyond the conventional ImageNet backbone. The framework adopts an encoder-decoder structure, with the encoder utilizing a Swin transformer to efficiently embed most important features. This process involves a Transfer Learning method, wherein layers from the Vision Transformer are converted by the Encoder Transformer and seamlessly integrated into a CNN Decoder. This methodology ensures minimal information loss from the original input image. The decoder employs a multi-decoding technique, utilizing dual decoders to generate two distinct attention maps. These maps are subsequently combined into a singular output via an additional CNN model. Our trained model MDS-ViTNet achieves state-of-the-art results across several benchmarks. Committed to fostering further collaboration, we intend to make our code, models, and datasets accessible to the public.

Coronary arterial dominance classification is essential for SYNTAX score estimation, which is a tool used to determine the complexity of coronary artery disease and guide patient selection toward optimal revascularization strategy. We developed coronary dominance classification algorithm based on the analysis of right coronary artery (RCA) angiograms using neural network.

We employed convolutional neural network ConvNext and Swin transformer for 2D image (frames) classification, along with a majority vote for cardio angiographic view classification. An auxiliary network was also used to detect irrelevant images which were then excluded from the data set.

5-fold cross validation gave the following dominance classification metrics (p = 95%): macro recall = 93.1% ± 4.3%, accuracy = 93.5% ± 3.8%, macro F1 = 89.2% ± 5.6%. The most common case in which the model regularly failed was RCA occlusion, as it requires utilization of left coronary artery (LCA) information.

The use of machine learning approaches to classify coronary dominance based on RCA alone has been shown to be successful with satisfactory accuracy. However, for higher accuracy, it is necessary to utilize LCA information in the case of an occluded RCA and detect cases where there is high uncertainty.

The loss landscape of neural networks is a critical aspect of their training, and understanding its properties is essential for improving their performance. In this paper, we investigate how the loss surface changes when the sample size increases, a previously unexplored issue. We theoretically analyze the convergence of the loss landscape in a fully connected neural network and derive upper bounds for the difference in loss function values when adding a new object to the sample. Our empirical study confirms these results on various datasets, demonstrating the convergence of the loss function surface for image classification tasks. Our findings provide insights into the local geometry of neural loss landscapes and have implications for the development of sample size determination techniques.

Efficiency and reliability optimization of distribution networks is an important task in the design of power supply systems, and its complexity increases with the development of new technologies such as distributed generation. One way to improve network reliability is through the installation and optimal placement of automatic circuit reclosers. The presence of distributed generation units and reclosers significantly increases the dimensionality of the optimization problem, thus necessitating the use of alternative approaches to solve it. The goal of the research is to analyze the effectiveness of metaheuristic algorithms in the recloser quantity and allocation optimization problem in a distribution network. The scientific novelty of the study lies in simultaneously considering the failure rate of network elements and changes in operating condition in case of contingencies. The practical significance of the work is demonstrated through the effectiveness of using metaheuristic methods when selecting the optimal equipment configuration in electrical networks. To solve the optimization problem of recloser placement in a 24-bus 10 kV network, the genetic algorithm, evolutionary strategy, and adaptive particle swarm optimization were considered. Computational experiments showed that the genetic algorithm is the most efficient in this case. The results can be further used in the development of methodological guidelines for designing distribution networks of various voltage classes.

LLM-based representation learning is widely used to build effective information retrieval systems, including scientific domains. For making science more open and affordable, it is important that these systems support multilingual (and cross-lingual) search and do not require significant computational power. To address this we propose SciRus-tiny, light multilingual encoder trained from scratch on 44 M abstracts (15B tokens) of research papers and then tuned in a contrastive manner using citation data. SciRus-tiny outperforms SciNCL, English-only SOTA-model for scientific texts, on 13/24 tasks, achieving SOTA on 7, from SciRepEval benchmark. Furthermore, SciRus-tiny is much more effective than SciNCL: it is almost 5x smaller (23 M parameters vs. 110 M), having approximately 2x smaller embeddings (312 vs. 768) and 2x bigger context length (1024 vs. 512). In addition to the tiny model, we also propose the SciRus-small (61 M parameters and 768 embeddings size), which is more powerful and can be used for complicated downstream tasks. We further study different ways of contrastive pre-training and demonstrate that almost SOTA results can be achieved without citation information, operating with only title-abstract pairs.

Time series representation learning is crucial in applications requiring sophisticated data analysis. In some areas, like the Oil and Gas industry, the problem is particularly challenging due to missing values and anomalous samples caused by sensor failures in highly complex manufacturing environments. Self-supervised learning is one of the most popular solutions for obtaining data representation. However, being either generative or contrastive, these methods suffer from the limited applicability of obtained embeddings, – so general usage is more often declared than achieved.

This study introduces and examines various generative self-supervised architectures for complex industrial time series. Moreover, we propose a new way to ensemble several generative approaches, leveraging the best advantages of each method. The suggested procedure is designed to tackle a wide range of scenarios with missing and multiscale data.

For numerical experiments, we use various-scale datasets of well logs from diverse oilfields. Evaluation includes change point detection, clustering, and transfer learning, with the last two problems being introduced for the first time. It shows that variational autoencoders excel in clustering, autoregressive models better detect change points, and the proposed ensemble succeeds in both tasks.

This paper examines changes in the content and technology of research in mathematics, computer science, and mathematical education determined by the digital transformation of civilization. Human activities and competences in the modern world are characterized in terms of an expanded personality. A characteristic feature here is the need and ability to solve fundamentally new problems, i.e., ones that are “not-known-how-to-solve.” The main types of human mathematical activities are classified. It is shown that, within the framework of school education, mathematics (including computer science) is an area where the task of developing the ability and readiness to solve challenging problems is solved most effectively. The possibilities of optimizing the educational process by using artificial intelligence tools are characterized.

This paper delves into the efficacy of utilizing the YOLOv8 model, which is based on a convolutional neural network (CNN), for the purpose of detecting objects within railway infrastructure, leveraging the capabilities of Neural Processing Units (NPU). It comprehensively explores various configurations of YOLOv8, each characterized by distinct architectural structures and input layer resolutions. These configurations were meticulously trained and evaluated using a sizable dataset comprising over 20   000 Full HD images. Through rigorous experimentation, this study elucidates the considerable potential of YOLOv8, especially when bolstered by NPU acceleration, in facilitating the real-time detection of objects within railway infrastructure. The performance of different YOLOv8 variants was thoroughly assessed by evaluating critical factors such as detection accuracy and computational efficiency. The findings of this research underscore the adaptability and resilience of YOLOv8 models across a spectrum of input resolutions, underscoring their proficiency in accurately identifying various elements of railway infrastructure under diverse environmental conditions. Furthermore, the integration of NPU acceleration emerges as a pivotal factor. It significantly augments the detection speed and responsiveness of the system, thereby enabling the swift processing of high-resolution images in real-time scenarios. This paper emphasizes the promising prospects associated with integrating YOLOv8 and NPU acceleration for applications in railway infrastructure monitoring and management. It offers valuable insights into the future trajectory of object detection technology within transportation systems, paving the way for enhanced efficiency and effectiveness in railway infrastructure operations.