Conference Papers (KIV)
Permanent URI for this collection
Browse
Recent Submissions
Item Enhancing Masked Language Modeling in BERT Models Using Pretrained Static Embeddings(Springer, 2026) Mištera, Adam; Král, PavelThis paper explores the integration of pretrained static fastText word vectors into a simplified Transformer-based model to improve its efficiency and accuracy. Despite the fact that these embeddings have been outperformed by large models based on the Transformer architecture, they can still contribute useful linguistic information, when combined with contextual models, especially in low resource or computationally constrained environments. We demonstrate this by incorporating static embeddings directly into our own BERTTINY-based models prior to pretraining using masked language modeling. In this paper, we train the models on seven different languages covering three distinct language families. The results show that the use of static fastText embeddings in these models not only improves convergence for all tested languages, but also significantly improves their evaluation accuracy.
Item Evaluating Feature Encodings for Unsupervised Machine Learning Classification in Automotive Ethernet Network(Institute of Electrical and Electronics Engineers, Inc., 2025) Anand, Kumar Ashutosh; Merz, Stefanie Angela; Heigl, Michael; Fiala, Dalibor; Schulz, Hannes; Kirmair, Wolfgang; Schramm, MartinCategorical attributes such as MAC and IP addresses constitute an integral part of Ethernet network data, and play a crucial role in modern network infrastructure. Representing these intrinsic entities with high cardinality presents a considerable performance challenge pertaining to machine learning tasks. In order to better manage the representations of the categorical attributes found in network data, this work presents new methods for transforming them. Some of these encoding schemes are designed using domain knowledge to limit the number of dimensions introduced in data while performing transformations. This study uses two specific Autoencoder deep neural networks for the unsupervised classification task to help assess the classification performance for the proposed encoding schemes. These varied encodings used to transform Ethernet network data from a real vehicle serve as a novel contribution to the feature engineering for analyzing the network data using machine learning approaches. The evaluation results show that the proposed techniques have a key impact on the classification performance, and the encoding schemes IE and ISF performed reasonably well in all three attack scenarios for each model.Item Large Language Models for Czech Aspect-Based Sentiment Analysis(Springer, 2026) Šmíd, Jakub; Přibáň, Pavel; Král, PavelAspect-based sentiment analysis (ABSA) is a fine-grained sentiment analysis task that aims to identify sentiment toward specific aspects of an entity. While large language models (LLMs) have shown strong performance in various natural language processing (NLP) tasks, their capabilities for Czech ABSA remain largely unexplored. In this work, we conduct a comprehensive evaluation of 19 LLMs of varying sizes and architectures on Czech ABSA, comparing their performance in zero-shot, few-shot, and fine-tuning scenarios. Our results show that small domain-specific models fine-tuned for ABSA outperform general purpose LLMs in zero-shot and few-shot settings, while fine-tuned LLMs achieve state-of-the-art results. We analyze how factors such as multilingualism, model size, and recency influence performance and present an error analysis highlighting key challenges, particularly in aspect term prediction. Our findings provide insights into the suitability of LLMs for Czech ABSA and offer guidance for future research in this area.Item BCI-Based Motor Imagery EEG Signal Classification Using a Novel Method (EEG-ITT) in Upper-Limb Exoskeleton(IEEE, 2024) Khoshkhooy Titkanlou, Maryam; Monjezi, Ehsan; Mouček, RomanBrain-computer interface (BCI) is an emerging technology that receives, processes, and converts brain signals into commands sent to output devices to perform desired tasks. Motor imagery (MI) based on electroencephalograms (EEGs) is one of the most widely used BCI paradigms, and it has demonstrated potential as an effective tool for neurorehabilitation. Recently, neural networks-in particular, deep architectures-have received substantial attention for the analysis of EEG signals (BCI applications). This paper proposes a new classification algorithm called EEG-ITT to increase the accuracy of classification motor imagery EEG signals using a non-invasive brain-computer interface. Utilizing a motor imagery dataset of 29 healthy subjects, including males aged 2126 and females aged 18-23, the proposed model demonstrated the highest accuracy, at 79.53 %. The noise injection method has also been implemented for data augmentation.Item Automatic detection of sleep spindles by neural networks algorithms(Czech Technical University, 2024) Rychlík, Jan; Mouček, RomanSleep constitutes an essential aspect of human existence, with the average individual dedicating approximately one-third of their life to this physiological activity. Consequently, comprehending and accurately analyzing sleep patterns is of paramount importance. This research aims to introduce, formulate, execute, and assess diverse machine/deep learning methodologies tailored for the processing of EEG signals geared explicitly towards identifying sleep spindles. The learning algorithms underwent training using meticulously annotated data from the Montreal Archive of Sleep Studies (MASS) data center. The convolutional neural network emerged as the most effective classification model, achieving an accuracy surpassing 67 %.
Item Enhanced Form-Based SPARQL Builder: Effortless Access to Temporal Medical Data for Users Without Technology Knowledge(IEEE, 2024) Včelák, Petr; Míka, Ondřej; Kryl, Martin; Klečková, JanaAccessing complex medical data, especially temporal information, presents a significant challenge for non-technical users, including healthcare professionals not versed in technology or query languages like SPARQL. This paper introduces an enhanced form-based SPARQL query builder that offers effortless access to temporal medical data. The described system simplifies constructing SPARQL queries into an intuitive, form-driven interface that abstracts and hides the technical complexities. Novel enhancement is primarily the user's ability to filter and order data with column constraints. By focusing on usability, the system empowers users to retrieve temporal medical information without needing in-depth knowledge of database systems or SPARQL syntax. We evaluate the system's usability and performance by testing it with non-technical healthcare professionals. Results demonstrate that users can easily navigate the interface, significantly reduce the time spent querying the database, and effectively retrieve data. This approach benefits from FAIR data principles and enhances data reusability in healthcare settings for reaching accuracy and precision for follow-up analysis in evidence-based research, facilitating better decision-making that can improve future patient outcomes.Item Automatic Motor Imagery Classification by CNN-Transformer-LSTM Using Multi-Channel EEG Signals(IOS Press, 2024) Pham, Duc Thien; Mouček, RomanThe brain-computer interface (BCI) is a promising technology that could bring about a significant revolution in various fields, including healthcare and human enhancement. One commonly used BCI method in healthcare, particularly in rehabilitation, is the analysis of motor imagery (MI) through an electroencephalogram (EEG). Our study introduces a hybrid deep learning model called CNN-Transformer-LSTM, which utilizes multi-channel EEG signals to classify MI binary and multiclass automatically. Our experiments have shown that this proposed method is more effective than previous state-of-the-art studies at accurately classifying MI using multi-channel EEG signals.Item Classification of MI EEG Signal Using Deep Learning Architectures for a Lower-Limb Rehabilitation Exoskeleton(Springer Nature Switzerland AG, 2025) Khoshkhooy Titkanlou, Maryam; Mouček, RomanRecent advances in neuroscience and engineering have resulted in brain-computer interface (BCI) devices that enhance the quality of life for people with movement limitations. BCI enables external devices to perform tasks using brain signals that are received, processed, and converted into commands by the brain. A widely used BCI paradigm based on electroencephalograms (EEGs) is motor imagery (MI), which has demonstrated potential as a tool for neurorehabilitation. In recent years, deep learning architectures have gained considerable attention for their ability to analyze EEG signals. This review paper focuses on applying deep learning for MI EEG classification in controlling lower-limb rehabilitation exoskeletons. Finally, current issues and potential directions will be discussed.Item Classification of MI EEG Signal Using Deep Learning Architectures for a Lower-Limb Rehabilitation Exoskeleton(Springer Nature Switzerland AG, 2025) Khoshkhooy Titkanlou, Maryam; Mouček, RomanRecent advances in neuroscience and engineering have resulted in brain-computer interface (BCI) devices that enhance the quality of life for people with movement limitations. BCI enables external devices to perform tasks using brain signals that are received, processed, and converted into commands by the brain. A widely used BCI paradigm based on electroencephalograms (EEGs) is motor imagery (MI), which has demonstrated potential as a tool for neurorehabilitation. In recent years, deep learning architectures have gained considerable attention for their ability to analyze EEG signals. This review paper focuses on applying deep learning for MI EEG classification in controlling lower-limb rehabilitation exoskeletons. Finally, current issues and potential directions will be discussed.Item A Type of EEG-ITNet for Motor Imagery EEG Signal Classification(SCITEPRESS – Science and Technology Publications, Lda, 2024) Khoshkhooy Titkanlou, Maryam; Monjezi, Ehsan; Mouček, RomanThe brain-computer interface (BCI) is an emerging technology that has the potential to revolutionize the world, with numerous applications ranging from healthcare to human augmentation. Electroencephalogram (EEG) motor imagery (MI) is among the most common BCI paradigms used extensively in healthcare applications such as rehabilitation. Recently, neural networks, particularly deep architectures, have received substantial attention for analyzing EEG signals (BCI applications). EEG-ITNet is a classification algorithm proposed to improve the classification accuracy of motor imagery EEG signals in a noninvasive brain-computer interface. The resulting EEG-ITNet classification accuracy and precision were 75.45% and 76.43%, using a motor imagery dataset of 29 healthy subjects, including males aged 21-26 and females aged 18-23. Three different methods have also been implemented to augment this dataset.Item Few-Shot Cross-Lingual Aspect-Based Sentiment Analysis with Sequence-to-Sequence Models(Springer, 2026) Šmíd, Jakub; Přibáň, Pavel; Král, PavelAspect-based sentiment analysis (ABSA) has received substantial attention in English, yet challenges remain for low-resource languages due to the scarcity of labelled data. Current cross-lingual ABSA approaches often rely on external translation tools and overlook the potential benefits of incorporating a small number of target language examples into training. In this paper, we evaluate the effect of adding few-shot target language examples to the training set across four ABSA tasks, six target languages, and two sequence-to-sequence models. We show that adding as few as ten target language examples significantly improves performance over zero-shot settings and achieves a similar effect to constrained decoding in reducing prediction errors. Furthermore, we demonstrate that combining 1,000 target language examples with English data can even surpass monolingual baselines. These findings offer practical insights for improving cross-lingual ABSA in low-resource and domain-specific settings, as obtaining ten high-quality annotated examples is both feasible and highly effective.Item Comparison of Road Traffic Division Methods for Distributed or Parallel Road Traffic Simulation(IEEE, 2025) Potužák, TomášIn this paper, the issue of difficult comparison of various road network division methods is addressed by designing a tool for comparison of multiple division methods in a unified environment. The functioning of this ROad Network DIvision BEnchmark Tool (RONDIBET) is demonstrated on a case study of comparison of two third-party division methods.Item Fitness-Fuction-Based Road Traffic Network Division(IEEE, 2025) Potužák, TomášIn this paper, a novel method for efficient division of road traffic network – the Multi-Level All-Possibilities-based Division (MLAPoD) – is described. Since a road traffic network is basically a graph, it is based on the standard multi-level graph partitioning. For the initial partitioning phase, it generates all possible assignments of graph nodes and assesses the quality of corresponding road traffic network divisions using a multi-objective fitness function. The obvious issue of a huge number of all possible assignments is solved by using sufficient number of graph coarsening steps to make the number of graph nodes feasible.Item Differential Equations as a Projection of Implicit Functions Using Spatio-Temporal Taylor Expansion and Critical Points Properties(AIP Publishing, 2024) Skala, VáclavThis contribution introduces a novel method for formulating differential equations. This method relies on expanding an implicit function that varies with time (denoted as "t") in the space-time domain using Taylor series. This formulation encompasses both ordinary differential equations (ODEs) and partial differential equations (PDEs).In the context of visualizing vector fields, such as fluid flow and electromagnetic fields, the critical points of ODEs play a crucial role in understanding physical phenomena behavior. This paper outlines a general approach for formulating ODEs and PDEs by treating them as time-varying scalar functions using the Taylor expansion. Furthermore, a new condition for identifying critical points is derived and specified specifically for cases where the function is invariant with respect to time (referred to as "t-invariant"). This newly derived formula enhances the detection of critical points, particularly in the context of acquiring and analyzing large 3D fluid flow data. This advancement enables efficient compression of 3D vector data and their representation through radial basis functions (RBFs).Item A New Projective Point in Convex Polygon Test with O(log N) Complexity(IEEE, 2024) Skala, VáclavThis contribution describes a novel and fully projective algorithm for a point-in-convex polygon test with computational complexity of O(log N) in E2 . The polygon vertices and tested points can be given in projective space, i.e., with the homogeneous coordinate w ̸= 0, without conversion to Euclidean space. It is independent of the polygon orientation. It uses vectorvector operations and is therefore aimed for implementation using SSE or AVX instructions. The algorithm is simple and robust, easy to implementItem TVMC: Time-Varying Mesh Compression Using Volume-Tracked Reference Meshes(Association for Computing Machinery, 2025) Chen, Guodong; Hácha, Filip; Váša, Libor; Dasari, MalleshamTime-varying meshes (TVMs), characterized by their varying connectivity and number of vertices, hold significant potential in AR/VR applications. However, their practical use is challenging due to their large file sizes and the complexity of time-varying topology. Many time-varying mesh compression methods attempted to exploit redundancy between consecutive meshes to compress TVMs more efficiently, however, most face difficulties in establishing stable vertex and surface correspondence between the frames of a TVM. We propose TVMC, a novel TVM compression method that leverages volume tracking and extracts high-quality reference meshes for inter-frame prediction. Specifically, we use as-rigid-as-possible volume tracking to align consecutive TVMs and track volume centers, followed by multidimensional scaling to refine reference centers. This allows us to precisely deform a group of frames to the reference space and extract the reference mesh which is then deformed to approximate each mesh in the group to get displacement fields for TVM compression. Extensive experiments show that TVMC outperforms state-of-the-art methods (e.g., Google Draco, V-DMC 4.0, etc.), with bitrates of 4-6 Mbps compared to 9-12 Mbps for Draco and 10-15 Mbps for V-DMC 4.0. It reduces the decoding time by 66.1% compared to Draco and enables an increased group of frames (up to 15) without significant distortion.Item Large Language Models for Summarizing Czech Historical Documents and Beyond(ScitePress, 2025) Tran, Václav; Šmíd, Jakub; Martínek, Jiří; Lenc, Ladislav; Král, PavelText summarization is the task of shortening a larger body of text into a concise version while retaining its essential meaning and key information. While summarization has been significantly explored in English and other high-resource languages, Czech text summarization, particularly for historical documents, remains underexplored due to linguistic complexities and a scarcity of annotated datasets. Large language models such as Mistral and mT5 have demonstrated excellent results on many natural language processing tasks and languages. Therefore, we employ these models for Czech summarization, resulting in two key contributions: (1) achieving new state-of-the-art results on the modern Czech summarization dataset SumeCzech using these advanced models, and (2) introducing a novel dataset called Posel od Čerchova for summarization of historical Czech documents with baseline results. Together, these contributions provide a great potential for advancing Czech text summarization and open new avenues for research in Czech historical text processing.Item Advancing Cross-Lingual Aspect-Based Sentiment Analysis with LLMs and Constrained Decoding for Sequence-to-Sequence Models(ScitePress, 2025) Šmíd, Jakub; Přibáň, Pavel; Král, PavelAspect-based sentiment analysis (ABSA) has made significant strides, yet challenges remain for low-resource languages due to the predominant focus on English. Current cross-lingual ABSA studies often centre on simpler tasks and rely heavily on external translation tools. In this paper, we present a novel sequence-to sequence method for compound ABSA tasks that eliminates the need for such tools. Our approach, which uses constrained decoding, improves cross-lingual ABSA performance by up to 10%. This method broadens the scope of cross-lingual ABSA, enabling it to handle more complex tasks and providing a practical, efficient alternative to translation-dependent techniques. Furthermore, we compare our approach with large language models (LLMs) and show that while fine-tuned multilingual LLMs can achieve comparable results, English centric LLMs struggle with these tasks.Item LACA: Improving Cross-lingual Aspect-Based Sentiment Analysis with LLM Data Augmentation(Association for Computational Linguistics, 2025) Šmíd, Jakub; Přibáň, Pavel; Král, PavelCross-lingual aspect-based sentiment analysis (ABSA) involves detailed sentiment analysis in a target language by transferring knowledge from a source language with available annotated data. Most existing methods depend heavily on often unreliable translation tools to bridge the language gap. In this paper, we propose a new approach that leverages a large language model (LLM) to generate high-quality pseudo-labelled data in the target language without the need for translation tools. First, the framework trains an ABSA model to obtain predictions for unlabelled target language data. Next, LLM is prompted to generate natural sentences that better represent these noisy predictions than the original text. The ABSA model is then further fine-tuned on the resulting pseudo-labelled dataset. We demonstrate the effectiveness of this method across six languages and five backbone models, surpassing previous state-of-the-art translation-based approaches. The proposed framework also supports generative models, and we show that fine-tuned LLMs outperform smaller multilingual models.Item Fast Approximate Symmetry Plane Computation as a Density Peak of Candidates(SciTePress, 2025) König, Alex; Váša, LiborSymmetry is a common characteristic exhibited by both natural and man-made objects. This property can be used in various applications in computer vision and computer graphics. There are various types of symmetries, amongst the most prominent belong reflection symmetries and rotation symmetries. In this paper, a method focusing on the fast detection of approximate reflection symmetry of a 3D point cloud with respect to a plane is proposed. The method is based on the creation of a set of candidates that are represented as rigid transformations, and have assigned weights, reflecting the estimated quality of the candidate. The final symmetry plane corresponds to a density peak in the transformation space. The method is demonstrated to be able to find symmetry planes in various objects in 3D, with its main benefit being the speed of the computation.
