Ponedeljkov seminar računalništva in informatike - Arhiv
| 2025 | 2024 | 2023 | 2022 | 2021 | 2020 | 2019 | 2018 | 2017 | 2016 | 2015 | 2014 | 2013 | 2012 |
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 |
sreda, 30. julij 2025 Matic POŽAR: Network-Aware Uplift Modeling through Influence Maximization Techniques in Dina MITROVIĆ: Analysis of full-length 16S rRNA gene and ITS regions amplicons sequenced with Oxford Nanopore Technologies
V ponedeljek, 4. avgusta 2025, bosta ob 16:00 uri izvedeni dve
predavanji v okviru PONEDELJKOVEGA SEMINARJA RAČUNALNIŠTVA IN INFORMATIKE
Oddelkov za Informacijske znanosti in tehnologije UP FAMNIT in UP IAM.
ČAS/PROSTOR: 4. avgust 2025 ob 16.00 prek Zoom-a (https://upr-si.zoom.us/j/297328207?pwd=S3Zpdk1VR3pjckNtWkQwKzlvcDR5UT09).
1. predavanje:
============
--------------------------------------
PREDAVATELJ: Matic POŽAR
--------------------------------------
Matic Požar is a 2nd year Data Science master's student at UP FAMNIT, writing his master thesis under the supervision of prof. Miklós Krész.
---------------------------------------------------------------------------------------------------------------------
NASLOV: Network-Aware Uplift Modeling through Influence Maximization Techniques
---------------------------------------------------------------------------------------------------------------------
POVZETEK:
Influence maximization is a fundamental problem in domains such as viral marketing, public health, and political campaigning, aiming to select a subset of users in a network to maximize the spread of influence. A central challenge in this task is the lack of knowledge about edge (infection) probabilities, which govern how influence propagates. We addresses this issue through the inverse infection problem: inferring edge probabilities from observed influence patterns. A novel diffusion and edge prediction model is proposed. Furthermore, we explore a novel application of this framework to uplift modeling, a technique used to identify individuals most likely to respond to interventions. By incorporating network effects into uplift modeling, our approach moves beyond the traditional assumption of user independence, enabling more realistic and effective targeting strategies in network settings. Leveraging recent advances in graph neural networks (GNNs), we propose a framework that captures complex network dependencies for accurate outcome estimation allowing for effective treatment allocation.
Seminar bo potekal v angleškem jeziku.
=============================================================================================================
2. predavanje:
============
---------------------------------------------
PREDAVATELJICA: Dina MITROVIĆ
---------------------------------------------
Dina Mitrović is a second-year Data Science master's student at UP FAMNIT and a participant in the pilot project GDI UP, currently writing her master's thesis under the supervision of Assist. Prof. Matjaž Hladnik, PhD.
--------------------------------------------------------------------------------------------------------------------------------------------------------------------
NASLOV: Analysis of full-length 16S rRNA gene and ITS regions amplicons sequenced with Oxford Nanopore Technologies
--------------------------------------------------------------------------------------------------------------------------------------------------------------------
POVZETEK:
This presentation introduces ongoing work aimed at proposing a standardized, open-source bioinformatics workflow for microbial community profiling using Oxford Nanopore Technologies (ONT) long-read amplicon sequencing. Unlike short-read methods, ONT enables full-length sequencing of 16S rRNA and ITS regions, offering the potential for improved species-level taxonomic resolution. However, reproducibility remains limited due to the lack of consistent analysis protocols. This research seeks to address that gap by evaluating existing bioinformatics tools and assembling a comprehensive pipeline that includes quality control, error correction, and taxonomic classification. Using sequencing data of mock microbial communities, this study will evaluate protocol performance and identify workflow steps that most influence classification accuracy. Key challenges to be addressed include sequencing error rates, primer and database biases, and tool compatibility. The ultimate goal is to improve accuracy, consistency, and reproducibility in long-read microbial profiling.
Seminar bo potekal v angleškem jeziku.
=============================================================================================================
Seminarja bosta izvedena online prek Zoom-a na sledeči povezavi:
https://upr-si.zoom.us/j/297328207?pwd=S3Zpdk1VR3pjckNtWkQwKzlvcDR5UT09
Meeting ID: 297 328 207
Passcode: 123456789
Vabljeni!
četrtek, 24. julij 2025 Nikola KOVAČEVIĆ: Discrete-Event Simulation Model for Wood Waste Reverse Supply Chain in Jovan PAVLOVIĆ Integrating agent-based modeling and network science for responsive epidemic management in urban transport
V ponedeljek, 28. julija 2025, bosta ob 16:00 uri izvedeni dve
predavanji v okviru PONEDELJKOVEGA SEMINARJA RAČUNALNIŠTVA IN INFORMATIKE
Oddelkov za Informacijske znanosti in tehnologije UP FAMNIT in UP IAM.
ČAS/PROSTOR: 28. julij 2025 ob 16.00 prek Zoom-a (https://upr-si.zoom.us/j/297328207?pwd=S3Zpdk1VR3pjckNtWkQwKzlvcDR5UT09).
1. predavanje:
============
---------------------------------------------
PREDAVATELJ: Nikola KOVAČEVIĆ
---------------------------------------------
Nikola Kovačević is 2nd year Data Science master's student at UP FAMNIT, writing his master thesis under the supervision of Prof. Balázs Dávid, PhD, and Co-Mentor: Črtomir Tavžes, PhD, Research Associate. He is also working at UP FAMNIT, as part of the pilot project GDI UP.
---------------------------------------------------------------------------------------------------------------
NASLOV: Discrete-Event Simulation Model for Wood Waste Reverse Supply Chain
---------------------------------------------------------------------------------------------------------------
POVZETEK:
Wood waste management faces significant challenges due to fragmented reverse supply chains and inconsistent data reporting systems. While traditional waste management approaches rely on linear disposal methods that limit resource recovery potential, reverse supply chains enable material valorization through collection, processing, and reuse pathways. To investigate modeling opportunities for real-world applications, we developed a discrete-event simulation model using Slovenia as a case study with data from the Slovenian Environment Agency (ARSO). The research addresses i) how fragmented wood waste data can be integrated into material flow understanding, ii) where bottlenecks exist in current systems, and iii) whether simulation modeling can provide insights for intervention strategies. The methodology combines data harmonization techniques with discrete-event simulation, tracking material flows from generation through collection to final processing, while implementing capacity constraints and operational dynamics. Initial results show the framework can pinpoint bottlenecks and evaluate scenarios for circular wood-waste reverse supply chains.
Seminar bo potekal v angleškem jeziku.
=============================================================================================================
2. predavanje:
============
------------------------------------------
PREDAVATELJ: Jovan PAVLOVIĆ
------------------------------------------
Jovan Pavlović is a second-year Data Science master's student at UP FAMNIT, under the supervision of mentor Miklós Krész and co-mentor László Hajdu. He is also part of the pilot project GDI UP.
----------------------------------------------------------------------------------------------------------------------------------------------------------------------
NASLOV: Integrating agent-based modeling and network science for responsive epidemic management in urban transport
----------------------------------------------------------------------------------------------------------------------------------------------------------------------
POVZETEK:
This presentation focuses on assessing different relative effectiveness of non-pharmaceutical epidemiological interventions through agent-based modeling. A schedule-based transit assignment model, FAST-TrIPs, is employed to realistically simulate passenger movements within the San Francisco Municipal Railway public transportation system, using accurate transit demand data. These simulations enable the construction of a detailed passenger encounter network that captures interactions relevant to disease transmission. From this network, another structure known as the vehicle trip network is constructed, modeling passenger transfer patterns. PageRank centrality analysis is applied to identify vehicle trips with the highest contagion potential, and the directed Louvain community-detection algorithm is separately used to identify vehicle trips bridging passenger communities. Building on insights from both analyses, a novel social-vehicle capacity restriction policy is proposed, in which capacity is selectively reduced on vehicle trips identified as most influential in disease spread. To evaluate this policy, a discrete compartmental Susceptible-Infected (SI) model is utilized to simulate infectious disease propagation across the passenger contact network. Primary finding indicate that measures reducing per‑contact transmission probability (e.g., mask wearing) dominate infection outcomes; even intelligently targeted capacity reductions play only a minor role compared with masking. However, when the a priori infection prevalence is high, more aggressive (and especially targeted) capacity limits can still shave off a few percentage points of spread.
Seminar bo potekal v angleškem jeziku.
=============================================================================================================
Seminarja bosta izvedena online prek Zoom-a na sledeči povezavi:
https://upr-si.zoom.us/j/297328207?pwd=S3Zpdk1VR3pjckNtWkQwKzlvcDR5UT09
Meeting ID: 297 328 207
Passcode: 123456789
Vabljeni!
ponedeljek, 21. julij 2025 Jakob BEBER: Naključnost v pokrivanju košev in Aljaž GEC: Generativna umetna inteligenca
V ponedeljek, 21. julija 2025, bosta ob 16:00 uri izvedeni dve
predavanji v okviru PONEDELJKOVEGA SEMINARJA RAČUNALNIŠTVA IN INFORMATIKE
Oddelkov za Informacijske znanosti in tehnologije UP FAMNIT in UP IAM.
ČAS/PROSTOR: 21. julij 2025 ob 16.00 prek Zoom-a (https://upr-si.zoom.us/j/297328207?pwd=S3Zpdk1VR3pjckNtWkQwKzlvcDR5UT09).
1. predavanje:
============
--------------------------------------
PREDAVATELJ: Jakob BEBER
--------------------------------------
Jakob Beber je študent magistrskega programa računalništvo in informatika. Na UP FAMNIT je dodiplomski študij zaključil z diplomsko nalogo o algoritmih, ki rešujejo problem pokrivanja košev. Raziskovanje tega problema nadaljuje na magistrskem študiju, zato bo na seminarju predstavil nove ugotovite.
--------------------------------------------------------
NASLOV: Naključnost v pokrivanju košev
--------------------------------------------------------
POVZETEK:
Na seminarju bo predstavljena vsebina članka, v katerem predstavimo sprotno različico problema pokrivanja košev in možne strategije za njegovo reševanje. Kasneje se osredotočimo na strategijo naslednji ustrezen (DNF), pri kateri ugotavljamo, da je analiza najslabšega primera preveč pesimistična za praktično uporabo. Da bi to raziskali, smo razvili spletno platformo za izvajanje in preizkušanje strategij ter različnih generatorjev zaporedij. V naših empiričnih rezultatih prikažemo razliko med teoretično najslabšim in "povprečnim" primerom uspešnosti za strategijo DNF.
Seminar bo potekal v slovenskem jeziku.
=============================================================================================================
2. predavanje:
============
----------------------------------
PREDAVATELJ: Aljaž GEC
----------------------------------
Aljaž Gec je magistrski študent računalništva in informatike na UP FAMNIT in zaposlen na srednji računalniški šoli kot učitelj strokovnih predmetov iz področja računalništva. Zanimajo ga področja umetne inteligence, programiranja in poučevanja le-tega. Njegov raziskovalni seminar je tesno povezan z njegovim delovnim področjem, saj obravnava razlago umetne inteligence pri generiranju različnih vsebin, med drugim tudi pisanju programov v različnih programskih jezikih.
---------------------------------------------------------
NASLOV: Generativna umetna inteligenca
---------------------------------------------------------
POVZETEK:
Generativna umetna inteligenca (UI) je vrsta UI, ki zna ustvarjati nove vsebine, kot so besedila, slike, koda, glasba in video. Deluje na podlagi učenja iz ogromnih količin podatkov in uporablja statistične vzorce za ustvarjanje novih rezultatov. Med glavnimi tehnologijami so transformerski modeli (npr. ChatGPT) in difuzijski modeli (npr. DALL·E, Stable Diffusion). V praksi lahko generativna UI npr. iz besedilnega opisa ustvari sliko ali napiše programsko kodo po navodilih. Uporablja se na različnih področjih, kot so umetnost, izobraževanje, programiranje in podpora strankam. V prihodnosti se pričakuje razvoj večmodalnih modelov, boljše etične smernice ter lokalno delovanje UI brez povezave v internet.
Seminar bo potekal v slovenskem jeziku.
=============================================================================================================
Seminarja bosta izvedena online prek Zoom-a na sledeči povezavi:
https://upr-si.zoom.us/j/297328207?pwd=S3Zpdk1VR3pjckNtWkQwKzlvcDR5UT09
Meeting ID: 297 328 207
Passcode: 123456789
četrtek, 10. julij 2025 Andrej ERJAVEC: Pristopi k iskanju informacij - od klasičnih do jezikovnih modelov in Marjan MEGLEN: Primerjava pristopov učenja pri matrični faktorizaciji v priporočilnih sistemih
V ponedeljek, 14. julija 2025, bosta ob 16:00 uri izvedeni dve predavanji v okviru PONEDELJKOVEGA SEMINARJA RAČUNALNIŠTVA IN INFORMATIKE Oddelkov za Informacijske znanosti in tehnologije UP FAMNIT in UP IAM.
ČAS/PROSTOR: 14. julij 2025 ob 16.00 prek Zoom-a (https://upr-si.zoom.us/j/297328207?pwd=S3Zpdk1VR3pjckNtWkQwKzlvcDR5UT09).
1. predavanje:
============
------------------------------------------
PREDAVATELJ: Andrej ERJAVEC
------------------------------------------
Andrej Erjavec je študent 2. letnika magistrskega študijskega programa Računalništvo in informatika na UP FAMNIT.
-----------------------------------------------------------------------------------------------------
NASLOV: Pristopi k iskanju informacij - od klasičnih do jezikovnih modelov
-----------------------------------------------------------------------------------------------------
POVZETEK:
Z naraščajočo količino digitalno shranjenih podatkov postaja učinkovito iskanje informacij eden osrednjih izzivov sodobne informacijske tehnologije. Področje iskanja informacij (Information Retrieval) se je oblikovalo kot odgovor na potrebo po sistematičnem dostopu do relevantnih vsebin v vse obsežnejših podatkovnih zbirkah. Poseben izziv predstavljajo nestrukturirani podatki, kot so besedila, slike in video posnetki, katerih delež v celotni količini podatkov že presega 80 odstotkov. V okviru tega seminarja bo predstavljen razvoj pristopov k iskanju informacij skupaj s konkretnimi implementacijami ter njihovimi področji uporabe.
Seminar bo potekal v slovenskem jeziku.
================================================================================
2. predavanje:
============
-------------------------------------------
PREDAVATELJ: Marjan MEGLEN
-------------------------------------------
Marjan Meglen je magistrski študent Računalništva in informatike na UP FAMNIT. Zanimajo ga področja strojnega učenja, priporočilnih sistemov in razlage modelov.
---------------------------------------------------------------------------------------------------------------------
NASLOV: Primerjava pristopov učenja pri matrični faktorizaciji v priporočilnih sistemih
---------------------------------------------------------------------------------------------------------------------
POVZETEK:
Na seminarju bomo predstavili matrično faktorizacijo kot enega najpogosteje uporabljenih pristopov sodelovalnega filtriranja v priporočilnih sistemih. Osredotočili smo se na primerjavo dveh pristopov učenja latentnih faktorjev: sočasnega in postopnega. Uporabili smo podatkovno zbirko Movielens, kjer smo metode primerjali glede na napovedno uspešnost in razložljivost modela. Za napovedno uspešnost smo uporabili metrikо RMSE, za razložljivost pa vizualizacije in statistične teste, kot sta Anova in parni t testi. Rezultati so pokazali, da je sočasno učenje omogočilo hitrejšo konvergenco in nižjo napovedno napako, medtem ko je postopno učenje zagotovilo boljšo razložljivost, predvsem pri filmih iz pogostejših žanrov. S tem smo pokazali, da izbira pristopa učenja pomembno vpliva na razmerje med napovedno natančnostjo in razložljivostjo modela.
Seminar bo potekal v slovenskem jeziku.
================================================================================
Seminarja bosta izvedena online prek Zoom-a na sledeči povezavi:
https://upr-si.zoom.us/j/297328207?pwd=S3Zpdk1VR3pjckNtWkQwKzlvcDR5UT09
Meeting ID: 297 328 207
Passcode: 123456789
ponedeljek, 7. julij 2025 Anđela ĐUKIĆ:Application of Clustering-Based Recommendation Systems in Nursing Diagnosis in Pika POVH MAVRIČ: Understanding Customer Behavior Through Data Analytics in the Banking Sector
V ponedeljek, 7. julija 2025, bosta ob 16:00 uri izvedeni dve
predavanji v okviru PONEDELJKOVEGA SEMINARJA RAČUNALNIŠTVA IN INFORMATIKE
Oddelkov za Informacijske znanosti in tehnologije UP FAMNIT in UP IAM.
ČAS/PROSTOR: 7. julij 2025 ob 16.00 prek Zoom-a (https://upr-si.zoom.us/j/297328207?pwd=S3Zpdk1VR3pjckNtWkQwKzlvcDR5UT09).
1. predavanje:
============
--------------------------------------------
PREDAVATELJICA: Anđela ĐUKIĆ
--------------------------------------------
Anđela Đjukić is a second year Data Science Masters student at UP FAMNIT under the supervision of mentor Bošjan Žvanut and co-mentor Branko Kavšek. She is also a part of the GDI UP Project at FAMNIT.
-------------------------------------------------------------------------------------------------------------------------
NASLOV: Application of Clustering-Based Recommendation Systems in Nursing Diagnosis
-------------------------------------------------------------------------------------------------------------------------
POVZETEK:
This work explores how recommendation systems, specifically the CLUICE algorithm, can help with more efficient and accurate nursing diagnoses for older adults in a shorter amount of time.
The process of making a nursing diagnosis involves many steps and a lot of time and effort, specifically when connecting symptoms to possible diagnoses, and there aren’t many tools that help with this in an automated way.
With this in mind, as well as the basic recommendation system knowledge, patients are modeled as users and their symptoms as items, using ideas from recommendation systems commonly seen and used in domains like marketing. The aim is to investigate if a system like this can suggest nursing diagnoses via identifying patterns in symptom data, even when the information is incomplete or limited.
The further goal comes through applying an approach not traditionally used in healthcare, with the potential to enhance clinical decision-making and reduce the cognitive burden on practitioners.
Seminar bo potekal v angleškem jeziku.
================================================================================
2. predavanje:
============
---------------------------------------------------
PREDAVATELJICA: Pika POVH MAVRIČ
---------------------------------------------------
Pika Povh Mavrič is currently completing her master’s studies in Computer Science at UP FAMNIT. She works as a data analyst in the Customer Relationship Management department at a bank, where she applies and deepens her expertise in data analytics. Her work involves automating processes, improving data quality, creating various reports and analyses, and applying data-driven decision-making to everyday processes within the department. The topic of her paper aligns well with her work in the banking sector.
--------------------------------------------------------------------------------------------------------------------------
NASLOV: Understanding Customer Behavior Through Data Analytics in the Banking Sector
--------------------------------------------------------------------------------------------------------------------------
POVZETEK:
This article examines how data analytics and machine learning are reshaping customer behavior analysis in the banking sector. It focuses on key applications such as credit scoring, fraud detection, churn prediction, and product recommendation, using models like Random Forest, LightGBM, and neural networks. The study highlights improved decision-making and customer engagement, while also addressing challenges around data quality, transparency, and ethics. It underscores the importance of organizational readiness and responsible AI practices to balance predictive accuracy with fairness and interpretability.
Seminar bo potekal v angleškem jeziku.
================================================================================
Seminarja bosta izvedena online prek Zoom-a na sledeči povezavi:
https://upr-si.zoom.us/j/297328207?pwd=S3Zpdk1VR3pjckNtWkQwKzlvcDR5UT09
