Wydział Informatyki
Loading...
Homepage
https://www.informatyka.agh.edu.pl/pl/
12 results
Search Results
Now showing 1 - 10 of 12
Item type:Article, Access status: Open Access , Configuring a hierarchical evolutionary strategy using exploratory landscape analysis(2023) Guzowski, Hubert; Smołka, Maciej
Wydział InformatykiHierarchic Memetic Strategy (HMS) is a stochastic global optimizer designed to tackle highly multimodal problems. It consists of parallel running optimization methods organized in a tree hierarchy. Depending on the task, different algorithms can be utilized on each of the levels. In this paper, we incorporate into HMS's structure a mechanism for choosing its configuration based on information gathered by a set of Exploratory Landscape Analysis (ELA) methods and hyperparametric optimization. We compared the performance of such configured HMS with a portfolio of proven state-of-the-art algorithms on the suite of black-box optimization functions. The results of this work show the efficacy of HMS and provide a set of default parameters evaluated for algorithms users. The use of ELA methods to select the configuration of a composite algorithm extends their standard use as part of an algorithm selector and provides insight into the relationship between exploration and exploitation for different types of fitness functions.Item type:Article, Access status: Open Access , Operating point shift induced by relay asymmetry: an iterative solution proposal(2023) Pekař, Libor; Matušů, Radek; Guzowski, Hubert; Gazdoš, František
Wydział InformatykiThis contribution focuses on a problem that appears when using a relay with non-symmetric output in the closed loop. Such a scheme is usually used for process model parameters identification, possibly followed by automatic controller tuning. Whenever static or dynamic properties of the process reveal asymmetry when the sign of the input changes, the setpoint (reference) becomes different from the operating point value of the process output. As a class of relay-based identification methods utilize calculations in the frequency domain that are based on integral computation around the operating point, the discrepancy between the setpoint and the operating point can lead to incorrect results. The aim of the paper is mainly to provide the reader with problem formulation and step-by-step proposition of how it can be solved. Concise numerical examples are also given. The concluding remarks suggest possible further ways of research.Item type:Doctoral Dissertation, Access status: Open Access , Organization and automation of a standards-aligned, declarative Network Slice/Subnet design process(Data obrony: 2025-12-08) Wyszkowski, Przemysław
Wydział InformatykiFinalnie, precyzowane są całościowe ramy proceduralne procesu projektowania, złożonego z wysokogranularnych aktywności o jasno określonym wejściu/wyjściu i komplementarnych wobec siebie rolach, wraz z określeniem stopniowych, kompatybilnych wstecznie możliwości wdrożenia automatyzacji projektowania NSI/NSSI w systemach zarządzania i orkiestracji sieci 5G. Opracowane koncepcje modeli i metod projektowania zostają poddane weryfikacji w części praktycznej dysertacji. Szeroki opis implementacji systemu opartego o proponowane ramy konceptualne, prezentuje ich realizowalność, wykazując jednocześnie szereg ich własności, takich jak spójność, wielowariantowość i rozszerzalność. Badania eksperymentalne zostały przeprowadzone na danych konfiguracyjnych zbliżonych do rzeczywistych, wykazując poprawność działania systemu w szeregu jednoczesnym braku konieczności zaangażowania operatora w proces projektowania i alokacji. Ewaluacja z perspektywy wydajnościowej wykazuje w pełni satysfakcjonujące czasy projektowania na bazie wejściowych wymagań klienta, pozwalając na zastosowanie przedstawionych rozwiązań dla dynamicznego dostarczania NSI/NSSI na żądanie, w dużych ilościach i w różnych skalach. Całość zgromadzonych rezultatów eksperymentalnych pozwala na potwierdzenie zasadności sformułowanej tezy, wykazując jednocześnie wysoki potencjał dla praktycznego wdrożenia opracowanych rozwiązań i ich standaryzacji.Item type:Book Chapter, Access status: Open Access , EXPBrain: Exponential Integrators for Glioblastoma Brain Tumor Simulations(Springer, 2025) Pabisz, Magdalena; Ciupek, Dominika; Vilkha, Askold; Paszyński, Maciej; Paszynski, M., Barnard, A.S., Zhang, Y.J. (eds)
Wydział InformatykiNOTE. This is a preprint of the paper with the same name in the Lecture Notes in Computer Science Journal. This preprint has not undergone peer review (when applicable) or any post-submission improvements or corrections. The Version of Record of this contribution is published in Paszynski, M., Barnard, A.S., Zhang, Y.J. (eds) Computational Science – ICCS 2025 Workshops. ICCS 2025. Lecture Notes in Computer Science, vol 15907, and is available online at: https://doi.org/10.1007/978-3-031-97554-7_10 In this paper we discuss a MATLAB implementation of the exponential integrators method employed for simulating of the brain tumor progression. As the input data we utilize publicly available T1-weighted magnetic resonance imaging dataset ds003826, representing healthy individuals. The data from these datasets are originally stored using NIfTI format. We select randomly one anonimized individual from the considered dataset. We normalize the brain scan data using min-max normalization to a range of 0 to 255. In the data from the dataset ds003826 the voxel resolution is not isotropic in all directions, so we interpolate the data from dimensions 176×248×256 into 194×248×256 in order to have proper proportions of the human brain. We set the data asa sequence of 256 PNG files with the resolution of 194 × 248. Having the MRI scan data, we run the exponential integrators method simulating the glioblastoma tumor growth using the Fisher-Kolmogorov diffusion-reaction model with logistic growth. We assume the initial tumor location and run the simulation predicting two years forward tumor growth. For the spatial discretization we employ the finite difference method, and for the temporal discretization we use the ultra-fast exponential integrators method. Our simulator generates the simulational results suitable for visualization using the ParaView tool.Item type:Doctoral Dissertation, Access status: Open Access , Objects pose tracking on RGB images(Data obrony: 2025-09-16) Majcher, Mateusz
Wydział InformatykiŚledzenie sześciowymiarowej pozy obiektu jest klasycznym problemem w widzeniu komputerowym, który jest wykorzystywany w wielu dziedzinach. Celem jest określenie trzywymiarowej rotacji oraz trzywymiarowej translacji względem obserwującej kamery. Ze względu na zmiany oświetlenia, przesłonięcia oraz niejednoznaczność widzianej strony obiektu, wyznaczanie pozy stanowi wymagające wyzwanie. Niniejsza rozprawa doktorska odnosi się do wybranych wyzwań, proponując metody wykorzystujące obrazy RGB, które poprawiają śledzenie pozy obiektu. Jednym z głównych wkładów jest wykorzystanie dodatkowych informacji z wcześniejszych klatek poprzez dostarczenie ich na wejście sieci neuronowej w celu poprawy precyzji określania punktów charakterystycznych w niejednoznacznych sytuacjach. W tym celu zaproponowane zostały nowe architektury sieci neuronowych, które dodatkowo posiadają osobno wytrenowaną część z wejściem do poprawy estymaty pozy obiektu na podstawie informacji o rotacji z poprzedniej klatki. Rozprawa bada metody do analizy zasłoniętych lub niewidocznych punktów charakterystycznych przy pomocy opisanych segmentów kształtu obiektu oraz odległości od rzutowanych krawędzi obiektu. Wyniki prac eksperymentalnych pokazują, że zaproponowane metody pozwalają w znacznym stopniu zniwelować negatywne wpływy przesłonięć oraz optymalnie wykorzystać istniejące dane w trakcie śledzenia pozy obiektów.Item type:Book Chapter, Access status: Open Access , Enhancing a hierarchical evolutionary strategy using the Nearest-Better Clustering(2024) Guzowski, Hubert; Smołka, Maciej; Pekař, Libor
Wydział InformatykiA straightforward way of solving global optimization problems is to find all local optima of the objective function. Therefore, the ability of detecting multiple local optima is a key feature of a practically usable global optimization method. One of such methods is a multi-population evolutionary strategy called the Hierarchic Memetic Strategy (HMS). Although HMS has already proven its global optimization capabilities there is an area for improvement. In this paper we show such an enhancement resulting from the application of the Nearest-Better Clustering. Results of experiments consisting both of curated benchmarks and a real-world inverse problem show that on average the performance is indeed improved compared to the baseline HMS and remains on par with state-of-the-art evolutionary global optimization methods.Item type:Article, Access status: Open Access , Maximizing efficiency: a comparative study of SOMA variants and constraint handling methods for time delay system optimization(2023) Senkerik, Roman; Kadavy, Tomas; Janku, Peter; Pluhacek, Michal; Guzowski, Hubert; Pekar, Libor; Matusu, Radek; Viktorin, Adam; Smołka, Maciej; Byrski, Aleksander; Komínková Oplatková, Zuzana
Wydział InformatykiThis paper presents an experimental study that compares four adaptive variants of the self-organizing migrating algorithm (SOMA). Each variant uses three different constraint handling methods for the optimization of a time delay system model. The paper emphasizes the importance of metaheuristic algorithms in control engineering for time-delayed systems to develop more effective and efficient control strategies and precise model identifications. The study includes a detailed description of the selected variants of the SOMA and the adaptive mechanisms used. A complex workflow of experiments is described, and the results and discussion are presented. The experimental results highlight the effectiveness of the SOMA variants with specific constraint handling methods for time delay system optimization. Overall, this study contributes to the understanding of the challenges and advantages of using metaheuristic algorithms in control engineering for time delay systems. The results provide valuable insights into the performance of the SOMA variants and can help guide the selection of appropriate constraint handling methods and the adaptive mechanisms of metaheuristics.Item type:Article, Access status: Open Access , Time-Frequency Token Advantage Clipping for Training Efficient Large Reasoning Model(2026) Bao, Rong; Wang, Bo; Li, Hongyu; Zheng, Riu; Wang, Xiao; Rutkowski, Leszek; Zhang, Qi; Ding, Liang; Tao, Dacheng
Wydział InformatykiLong Chain-of-Thought (CoT) reasoning enhances large reasoning models’ performance but suffers from severe inefficiencies, as models often overthink simple problems or underthink complex ones. Current sequence-level optimizations, like length penalties, are too coarse-grained to distinguish core logic from verbose language, precluding the necessary token-level control for efficient reasoning CoT. To overcome these limitations, we introduce Time-Frequency token Advantage Clipping (TFAC), a novel training framework designed to build efficient large reasoning models via token-level interventions. Specifically, TFAC functions along two dimensions: 1) The Frequency Dimension: It discourages inefficient loops and encourages deeper exploration by dynamically reducing the advantage scores of high-entropy tokens that are repeatedly generated within a single reasoning path. 2) The Time Dimension: It reduces excessive overthinking of the system by establishing a historical baseline for the occurrence count of each critical token in previously successful trajectories, and clipping the advantages of tokens that exceed this baseline during training. Crucially, to preserve the model’s exploratory capabilities on novel problems, this suppression mechanism is automatically disabled when no historical record of success is available. Experiments conducted on the Deepseek-Distill-32B and Qwen3- 8B models show that TFAC outperforms leading baseline methods, improving performance by 2.3 and 3.1 percentage points, respectively, while simultaneously reducing inference costs by 35% and 28% in scenarios where correct answers are generated. These results validate the significant efficacy of TFAC in training large reasoning models that are both powerful and highly efficient. The source code and datasets used in this study are available at https://github.com/rbao2018/TFAC.Item type:Article, Access status: Open Access , Bridging the Tokenizer Gap: Semantics and Distribution-aware Knowledge Transfer for Unbiased Cross-Tokenizer Distillation(2026) Wang, Huazheng; Jing, Yongcheng; Sun, Haifeng; Wang, Jingyu; Liao, Jianxin; Rutkowski, Leszek; Tao, Dacheng
Wydział InformatykiCross-tokenizer knowledge distillation, where the teacher and student employ different tokenizers, is becoming increasingly prevalent, yet it poses underexplored challenges: existing methods fail to capture the rich knowledge encoded in teacher logits, as evidenced by the neglect of semantic information, inaccurate and biased logit alignment, and discarding distributional structure—ultimately leading to unfavorable distillation. To address these issues, we propose SEDI, a semantics and distribution-aware knowledge transfer framework tailored for cross-tokenizer distillation. To preserve factual knowledge, SEDI employs bipartite graph-based alignment at the tokenization level and a sliding window re-encoding strategy at the vocabulary level, enabling unbiased transfer of the teacher’s next-token predictions into the student’s vocabulary space. To further retain distributional information, we align the student’s entropy with that of the teacher by incorporating the student’s own logits during training, which helps to mitigate the exposure bias problem. Experiments on ten datasets across three task domains and five different teacher-student model pairs with varying vocabulary sizes demonstrate that SEDI delivers substantial improvements, with gains of up to 19.8 .Item type:Article, Access status: Open Access , Efficient time-delay system optimization with auto-configured metaheuristics(2024) Senkerik, Roman; Guzowski, Hubert; Janku, Peter; Kadavy, Tomas; Kominkova Oplatkova, Zuzana; Matusu, Radek; Pluhacek, Michal; Pekar, Libor; Viktorin, Adam; Byrski, Aleksander; Smołka, Maciej
Wydział InformatykiThis paper presents an experimental study that compares the performance of four selected metaheuristic algorithms for optimizing a time delay system model. Time delay system models are complex and challenging to optimize due to their inherent characteristics, such as non-linearity, multi-modality, and constraints. The study includes an explanation of the choice and core functionality of the selected algorithms, which are both baseline and state-of-the-art variants of self-organizing migrating algorithm (SOMA), state-of-the-art variant from the Success-History-based Adaptive Differential Evolution family of algorithms, with emphasis on diverse search (DISH algorithm), and Covariance Matrix Adaptation Evolution Strategy (CMA-ES) algorithm. The hyperparameters of the metaheuristic algorithms were set using the iRace automatic algorithm configuration framework. The paper emphasizes the importance of metaheuristic algorithms in control engineering for time-delay systems to develop more effective and efficient control strategies and precise model identifications. The experimental results highlight the effectiveness of the state-of-the-art algorithms with specific adaptive mechanisms like population organization process, diverse search and adaptation mechanisms ensuring a gradual transition from exploration to exploitation. Overall, this study contributes to understanding the challenges and advantages of using metaheuristic algorithms in control engineering for time delay systems. The results provide valuable insights into the performance of modern metaheuristic algorithms and can help guide the selection of appropriate adaptive mechanisms of metaheuristics.