Planiranje kretanja robota je jedan od fundamentalnih problema u robotici. Cilj klasičnog problema planiranja kretanja je pronaći trajektoriju koju robot treba da prati da bi izvršio zadatak, pri čemu mora izbjegavati koliziju sa preprekama i ostalim sudionicima. Problem postaje dodatno izazovan ako algoritam mora odrediti i sekvencu zadataka (engl. Task and Motion Planning - TAMP), npr. otvoriti frizider, izvaditi teglu, itd.

The task of forestry companies is to contribute, together with representatives of non-governmental organizations and experts from the academic community, to knowledge about how climate change in Bosnia and Herzegovina affects forestry as an industrial branch, but also biodiversity. Sustainable management of forest ecosystems is a priority for local communities, administrative units, and the overall state strategy of each country. In Bosnia and Herzegovina, according to the latest Large-Scale Forest Inventory (Lojo et al., 2008), natural forests occur on over 50 percent of the total territory of the country. The survival of forest tree species is closely linked to the survival of wild animal species. Each statistical contribution to the knowledge on individual species behavior in the conservation context is valuable and desirable. In terms of exploring the existence of drought-resistant individuals in our ecosystems, the study was aimed.

U ovom projektu studentice i studenti će raditi na razvoju i optimizaciji metoda vještačke inteligencije (AI) za analizu i rekonstrukciju putanje rasutih gama zraka u detektorima visokih energija. Inspiracija dolazi iz postojećih rješenja opisanih u radovima gdje se koriste duboke neuronske mreže i probabilistički pristupi za prepoznavanje elektronskih tragova pri niskim energijama.

Cilj ovog istraživačkog projekta je da se kroz multidisciplinarni pristup razviju i testiraju poboljšanja u dizajnu sistema Expert Experiments s fokusom na korisnički interfejs, prikupljanje podataka i primjenjivost u formalnim i neformalnim obrazovnim okruženjima (škole, kampovi, online kursevi). Pored toga, studenti će koristiti ovu platformu za prikupljanje podataka iz stvarnog okruženja (npr. temperatura tla, promjene u pritisku, detekcija kretanja) i evaluirati je kroz eksperimente sa srednjoškolcima i studentima.

U ovom projektu studentice i studenti će se upoznati sa najnovijim izdanjem COSI Data Challengea (DC3), koje simulira podatke iz Compton teleskopa COSI planiranog za lansiranje 2027. godine. Fokus je na instalaciji i korištenju softverskog paketa cosipy (kao dijela šireg alata COSITools), analizi FITS fajlova sa spektralnim i slikovnim informacijama te svladavanju tehnika za procjenu pozadinske emisije (kontinuum i linijska emisija).

Cilj ovog istraživanja je sistematski analizirati sve Bayesian optimizacijske algoritme (HPO metode) koji su testirali LassoBench, kako bi se utvrdilo koji metod najbolje funkcionira u kontekstu Weighted Lasso regresije.

Periphytic algae are essential components of aquatic ecosystems and have gained growing interest in forensic science due to their potential as biological tracers in investigations. This study explores the colonization dynamics of periphytic algae on various synthetic and natural substrates over a 56-day period in a natural freshwater environment. Substrates including cotton, nylon, leather, metal, plastic, glass, and wood were submerged to assess differences in algal succession and biofilm formation.

Nanotehnologija omogućava razvoj novih tipova senzora sa izuzetnom preciznošću, koji se koriste u širokom spektru uređaja, od pametnih telefona do medicinskih aparata. Ovaj projekat istražuje kako nanotehnologija omogućava miniaturizaciju senzora i poboljšanje njihovih performansi.

Image classification is an important task in computer vision field. However, current image classification models are not human understandable. The goal of this research is to achieve explainability via diffusion based concept swapping.

Od otkrića cjelobrojnog kvantnog Hall-ovog efekta 1980. godine, topološke faze materije postale su jedno od najistraživanijih oblasti u modernoj fizici čvrstog stanja. Ovi mezoskopski sistemi se mogu jednostavno opisati topološkom invarijantom - fizikalnom veličinom koja se može izračunati iz sopstvenih stanja Hamiltonijana koji opisuje sistem u impulsnom prostoru. Topološka invarijanta uzima cjelobrojne vrijednosti i odgovorna je za postojanje vezanih stanja na granicama kvantnog sistema.

Pruned modeli za obradu slika se široko koriste u različitim aplikacijama gdje su efikasnost, brzina i manja potrošnja memorije kritični npr. mobilni, IoT, medicinski uređaji itd.

Primjena dubokog učenja za precizno određivanje spola u forenzičkoj odontologiji: Studija na dječjoj i adolescentskoj populaciji Identifikacija spola je ključna u forenzičkim istragama, posebno u slučajevima koji uključuju nepoznate osobe, nestale osobe ili žrtve gnusnih zločina. Neke studije sugeriršu da mogu postojati razlike u vremenu i brzini razvoja trećeg molara između muškaraca i žena.

Brzine kretanja u krajevima bliskih galaksija kao moguć problem za teorije tamne materije Duraković Amel i Mistele Tobias Observatoire astronomique de Strasbourg, Institute of Physics of the Czech Academy of Sciences Case Western Reserve University

Rješavanje Poissonove jednačine i njene modifikacije u galaktičkim okolnostima Duraković Amel i Mistele Tobias Observatoire astronomique de Strasbourg, Institute of Physics of the Czech Academy of Sciences Case Western Reserve University

Health data are characterised by the large numbers of variables categorical in their nature. One-hot encoding is commonly used in health data analysis when dealing with categorical variables. It is a popular technique used in machine learning to represent categorical variables numerically. It converts categorical variables into binary vectors, where each category is represented by a binary value (0 or 1) in a separate feature. While one-hot encoding can be useful in certain scenarios, it can also have an impact on model performance e.g., exacerbate the curse of dimensionality problem and model complexity and interpretability. This project aims to develop novel encoding method using real health data and compare it against the conventional such as one-hot encoding and embedding-based.

Cilj ovog projekta je unapređenje postoječih algoritama planiranja sa metodama mašinskog učenja da bi brže pronalazili validne planove u sličnim situacijama korištenjem prethodnog iskustva.

Ciljani projekat ucenja analiziranja relevantnih clanaka iz oblasti sumarstva, s pracenjem ostalih izvora informacija o izvrsenju kriminaliteta unutar sumskih gazdinstava, lovista, kao i izvora informacija o kradji lovackih pasa, ilegalnoj distribuciji lovnih dobara, lovnih pasa i organizaciji drugih ilegalnih aktivnosti u sumama u BiH.

In this project, we are looking to apply Codon and its NumPy libraries to implement various LLMs, such as Gemini, Llama, Mixtral, and to compare the new implementations with the current state-of-the-art implementations in other languages.

In this project, we are looking to compare this new implementation with the reference NumPy implementation, as well as the other performant NumPy implementations (e.g., Cython, Numba, Pythran, etc.) through the standardized benchmark sets such as NPBench.

This project aims to study the rotation and total magnetization (as a function of time) of spherical, cubic, and disk-shaped, multidomain magnetic colloids in oscillating magnetic fields. To simulate the internal magnetic structure and dynamics of the colloids, we will take a multiscale approach consisting of a single-spin Landau-Lifshitz-Gilbert equation model [3], and a raspberry colloid model where the domains are thermal Stoner-Wohlfarth particles [4]. The knowledge gained in contrasting these approaches to “ground-truth” micromagnetics, using varying levels of complexity, will circumvent current limitations and help expand the potential applications of magnetic fluids in cancer treatment.

Kandidati/Kandidatkinje će zajedno sa mentorom analizirati druge kanale produkcije i raspada ovog leptokvarka. Tačnije, analizira ćemo druge moguće kombinacije produkcije i raspada koje uključuju kvarkove i leptone prve i druge generacije (odnosno kombinacije down-quarka/strangequarka sa elektronima ili mionima). Konačni rezultati bi trebali biti granice na parametarski prostor leptokvarka, odnosno na njegovu masu i na njegovu konstantu vezivanja). Naposljetku ćemo uporediti izvedene rezultate sa granicama koje proizilaze iz eksperimenata na niskim energijama.

This project aims to conduct a comprehensive comparative analysis of various state-of-the-art path planning algorithms, including A*, Theta*, TO-AA (TO-AA-SIPP), Anya, Polyanya, Informed RRT*, BIT*, RRT, and RRT*. By evaluating the behavior and performance of these diverse navigation approaches, the project seeks to identify their strengths, weaknesses, and optimal application scenarios in dynamic and static environments. The ultimate goal is to provide insights that could guide the selection and implementation of path-planning algorithms in robotics and autonomous systems across various domains.

Local explanation methods are essential in explainable artificial intelligence (XAI), providing insights into the decision-making processes of machine learning models at the instance level. These methods offer transparent, understandable explanations for specific predictions made by complex models, such as deep neural networks or ensemble methods. This comparative analysis evaluates and contrasts various local explanation techniques to assess their effectiveness, interpretability, and applicability across different domains and model architectures.

This project idea revolves around creating a sophisticated and intuitive interface engineered to deliver multimodal explanations suitable for various data types including tabular, textual, and visual data. The interface should be designed to incorporate a wide array of explanation methods, tailored to meet the specific interpretability requirements of different users. By integrating diverse interpretable models and techniques, the interface facilitates a comprehensive understanding of the decision-making processes underpinning machine learning models employed across multiple domains.