Bulletin of Natural Sciences Research
https://aseestant.ceon.rs/index.php/bnsr
Faculty of Sciences and Mathematics, University of Priština in Kosovska Mitrovica, Serbiaen-USBulletin of Natural Sciences Research 2738-0971<p>Authors retain copyright and grant the journal right of first publication with the work simultaneously licensed under a <a href="http://creativecommons.org/licenses/by/3.0/" target="_new">Creative Commons Attribution License</a> that allows others to share the work with an acknowledgement of the work's authorship and initial publication in this journal.</p>OPTIMIZATION OF TOKENIZATION AND MEMORY MANAGEMENT FOR PROCESSING LARGE TEXTUAL CORPORA IN MULTILINGUAL APPLICATIONS
https://aseestant.ceon.rs/index.php/bnsr/article/view/54993
<p class="MsoNormal" style="text-align: justify; margin: 0cm 18.45pt .0001pt 14.2pt;"><strong><span style="font-size: 10.0pt; line-height: 107%; font-family: 'Times New Roman','serif';">Optimization of tokenization and memory management in processing large datasets represents a key challenge in the contemporary development of language models. This paper focuses on enhancing the processing of large textual corpora in Serbian using the GPT-2 model, specifically adapted for transfer learning. Tokenization optimization was achieved by adding language-specific tokens for Serbian, while memory management was improved through advanced resource management methods during training. Key findings demonstrate significant memory consumption reduction and training process acceleration, enabling more efficient utilization of available computational resources. This research contributes to the development of language models tailored for the Serbian language and provides a foundation for further studies in the field of natural language processing (NLP). The implications of this work are multifaceted: it facilitates more efficient creation of NLP applications for Serbian-speaking regions, enhances the accuracy and performance of language models, and opens opportunities for applications across various domains, from automated translation to sentiment analysis. This study paves the way for future research focusing on additional optimization of language models, including adaptation for other languages with similar characteristics, as well as exploring new methods for even more efficient memory management during large-scale textual data processing.</span></strong></p>Dejan DodićDušan RegodićNikola Milutinović
Copyright (c) 2025 Bulletin of Natural Sciences Research
2025-02-062025-02-0615110.5937/bnsr15-54993EXPLORING THE NEXUS OF TOURISM DEVELOPMENT, COMMUNITY PERCEPTIONS, AND SUSTAINABILITY IN PROTECTED AREAS
https://aseestant.ceon.rs/index.php/bnsr/article/view/55046
<p class="MsoNormal" style="text-align: justify; margin: 0cm 18.5pt .0001pt 14.2pt;"><strong style="mso-bidi-font-weight: normal;"><span style="font-size: 10.0pt; line-height: 107%; font-family: 'Times New Roman','serif';">Sustainable tourism integrates economic, social, and environmental aspects of sustainability. This study investigates the local community’s perceptions of tourism development impacts and factors influencing support for sustainable tourism and destination sustainability within Stara Planina Nature Park. Using a Structural Equation Modeling (SEM) approach, using multiple hypothesized relationships across key dimensions, including economic, environmental, social, and infrastructural impacts are examined. The findings highlight the importance of socio-cultural factors in fostering support, while also recognizing the negative impact of environmental and infrastructural concerns. Socio-cultural impacts significantly and positively influenced support for sustainable tourism and destination initiatives, highlighting the role of cultural exchange, tradition preservation, and community identity in garnering local support. These findings align with previous studies, emphasizing the importance of perceived socio-cultural benefits in fostering community backing for tourism development. Effective STD management requires the active involvement of local stakeholders to ensure alignment with local values and environmental goals. Policymakers should focus on enhancing socio-cultural benefits, addressing infrastructural challenges, and effectively communicating economic advantages. Limitations of the study include its cross-sectional design, suggesting the need for longitudinal research to better understand the evolving impact of tourism.</span></strong></p>Sanja Obradović Strålman Nikola Milentijevic
Copyright (c) 2025 Bulletin of Natural Sciences Research
2025-02-062025-02-0615110.5937/bnsr15-55046OPTIMIZATION OF FLUID VOLUME CONTROL IN HEMODIALYSIS USING FEDERATED LEARNING
https://aseestant.ceon.rs/index.php/bnsr/article/view/55563
<p><strong><span style="font-size: 11.0pt; mso-bidi-font-size: 10.0pt; line-height: 107%; font-family: 'Calibri','sans-serif'; mso-fareast-font-family: Calibri; mso-bidi-font-family: 'Times New Roman'; mso-ansi-language: EN-US; mso-fareast-language: EN-US; mso-bidi-language: AR-SA;">Overhydration (OH) represents a significant challenge for hemodialysis patients, significantly affecting the outcomes of their treatment. Accurate prediction and management of overhydration are key to optimizing therapy and improving patients' quality of life. The aim of this paper is to present a federated learning (FL)-based approach designed to predict overhydration in hemodialysis patients, using a dataset comprising different clinical and bioimpedance parameters. Federated learning enables collaborative learning from multiple data sources while preserving the privacy and security of individual patient data. Research results show that federated learning has the potential as an effective tool for predictive modeling in clinical settings. The developed models achieve high performance in overhydration estimation, with metrics confirming their accuracy and reliability. The proposed approach achieved a R² of </span><span lang="SR" style="font-size: 11.0pt; mso-bidi-font-size: 10.0pt; line-height: 107%; font-family: 'Calibri','sans-serif'; mso-fareast-font-family: Calibri; mso-bidi-font-family: 'Times New Roman'; mso-ansi-language: SR; mso-fareast-language: EN-US; mso-bidi-language: AR-SA; mso-bidi-font-weight: bold;">0.9999999, a MAE of </span><span lang="sr-Latn-RS" style="font-size: 11.0pt; line-height: 107%; font-family: 'Calibri','sans-serif'; mso-fareast-font-family: Calibri; mso-bidi-font-family: 'Times New Roman'; mso-ansi-language: #241A; mso-fareast-language: EN-US; mso-bidi-language: AR-SA;">0.00018</span> <span style="font-size: 11.0pt; mso-bidi-font-size: 10.0pt; line-height: 107%; font-family: 'Calibri','sans-serif'; mso-fareast-font-family: Calibri; mso-bidi-font-family: 'Times New Roman'; mso-ansi-language: EN-US; mso-fareast-language: EN-US; mso-bidi-language: AR-SA;">and an MSE of </span><span lang="sr-Latn-RS" style="font-size: 11.0pt; line-height: 107%; font-family: 'Calibri','sans-serif'; mso-fareast-font-family: Calibri; mso-bidi-font-family: 'Times New Roman'; mso-ansi-language: #241A; mso-fareast-language: EN-US; mso-bidi-language: AR-SA;">0.0031</span><span style="font-size: 11.0pt; mso-bidi-font-size: 10.0pt; line-height: 107%; font-family: 'Calibri','sans-serif'; mso-fareast-font-family: Calibri; mso-bidi-font-family: 'Times New Roman'; mso-ansi-language: EN-US; mso-fareast-language: EN-US; mso-bidi-language: AR-SA;">, demonstrating its predictive strength and practical applicability. This study highlights the advantages of federated learning in using distributed data to advance predictive capabilities in healthcare. By overcoming challenges related to privacy and data security, the approach presented in this paper opens up opportunities for more personalized and accurate prognoses, potentially improving decision-making and patient care in hemodialysis.</span></strong></p>Suzana ĐorđevićStefan ĆirkovićDanijela MiloševićMilan GligorijevićVladimir Mladenović
Copyright (c) 2025 Bulletin of Natural Sciences Research
2025-03-022025-03-0215110.5937/bnsr15-55563