PLAN RADA SEMINARA ZA NOVEMBAR 2021. GODINE
Zbog trenutne epidemiološke situacije, predavanja na seminaru će se održavati na daljinu, a slušaoci mogu da ih prate preko linka https://miteam.mi.sanu.ac.rs/asset/YoqHWKALRkRTbK9So.
Registracija za on-line praćenje predavanja na Seminaru je na linku https://miteam.mi.sanu.ac.rs/asset/xzGqvSp7aWbg8WpYX.
Utorak, 02.11.2021. u 14:15, Online
Vladan Radosavljević, Machine Learning Chapter Lead at Spotify
BEHIND-THE-SCENES OF HOMEPAGE PERSONALIZATION AT SPOTIFY
Hundreds of millions of users listen to audio content on Spotify each month, with more than fifty million daily active users listening the content from the homepage alone. On the homepage, we aim to surface the best of Spotify for each user by providing a personalized space where users can find recommendations of playlists, albums, artists, podcasts and more, all tailored to users’ individual preferences. In this talk, we will explain the methods and technologies that are used in the end-to-end process of homepage personalization. In addition, we will present some of the challenges we faced in implementing such solutions in a production environment at scale and the approaches we developed to address them.
Petak, 05.11.2021. u 12:00, Online
Luka Matijević, Matematički institut SANU
METAHEURISTIC APPROACHES FOR THE GREEN VEHICLE ROUTING PROBLEM
Vehicle routing problem (VRP) belongs to the class of optimization problems that has many real-life applications. Usually, the objective is to minimize the distance, time, or travel costs. Recently, the green vehicle routing problem (GVRP) gained much attention, because it aims to reduce the environmental impact of the engaged vehicles. The main goal is to minimize greenhouse gasses emissions (GHG) produced by a fleet of vehicles. Both internal combustion vehicles (ICV) and alternative fuel vehicles (AFV) are considered, dividing GVRP into two separate subclasses: ICV-based GVRP and AFV-based GVRP. In the ICV-based subclass, the environmental aspect comes from the objective function which aims to minimize GHG emissions or fuel usage of ICVs. This is usually connected to the minimization of travel time, instead of the distance. Busy city streets are avoided and replaced by the fast autoroutes whenever is possible. On the other hand, the environmental aspect of AFV-based GVRP is implicit and comes from avoiding fossil fuels in transport. Since GVRP is NP-hard, finding the exact solution in a reasonable amount of time is often impossible for larger instances, which is why metaheuristic approaches are predominantly used. In the field of optimization, metaheuristics are high-level methods that guide some underlying heuristic, in order to improve its performance. The focus of this talk will be to introduce the problem, present some of the most common GVRP attributes, and review the existing literature applying metaheuristic approaches to this problem. Finally, we are going to discuss potential directions for future research.
Zajednički sastanak sa Studentskim seminarom.
Utorak, 09.11.2021. u 14:15, Live stream Niš
Ivan Milenković, Fakultet organizacionih nauka
OKVIR ZA EVALUACIJU MULTIMODALNIH BIOMETRIJSKIH SISTEMA
Primena biometrijskih tehnologija danas u ljudskom društvu postaje sve češća, gotovo da možemo konstatovati da je ona deo naše svakodnevnice. Za odabir odgovarajućeg biometrijskog modaliteta, kao i algoritama za rad sa biometrijskim modalitetom, neophodno je sprovesti odgovarajuću evaluaciju performansi rada biometrijskog sistema. Ipak, ovu evaluaciju nije uvek lako sprovesti. Radi prevazilaženja ovih teškoća, definisan je objedinjeni model evaluacije multimodalnih biometrijskih sistema. U okviru objedinjenog modela dat je metamodel evaluacije multimodalnih biometrijskih sistema, koji predstavlja svojevrsnu ontologiju pojmova značajnih za ovu oblast. Primenom ovog metamodela, moguće je kreirati modele evaluacije različitih biometrijskih sistema. Na osnovu modela evaluacije multimodalnih biometrijskih sistema kreiran je prototip okvira za evaluaciju multimodalnih biometrijskih sistema. Razvijena je i nova metoda za analitičko određivanje praga osetljivosti u skladu sa postavljenim parametrima željenog ponašanja sistema, kao i alat baziran na funkcionalnostima okvira koji ima za cilj poboljšanje efikasnosti procesa obrazovanja inženjera u oblasti biometrije.
Utorak, 16.11.2021. u 14:15, Online
Una Stanković, Matematički institut SANU
THE WEIGHTED SCHEDULING PROBLEM WITH DEADLINES AND RELEASE TIMES
We consider a non-preemptive scheduling problem with deadlines and release times, where each task is associated with a nonnegative weight. The problem consists of finding a subset of tasks that are going to be scheduled and executed in such a way that the sum of weights of all scheduled tasks is maximized. This problem has many potential applications, one of them in energy efficient BlockChain systems, the ones that use Proof-of-Useful-Work (PoUW) consensus protocols. For the considered problem, we propose two mathematical formulations in the form of Mixed Integer Linear Programs, and test them on a set of randomly generated instances using GLPK solver. Since the problem is proven to be NP-hard, we consider a metaheuristic approach, based on the General Variable Neighborhood Search (GVNS) method. This approach is tested on the same set of randomly generated instances, and compared with the exact solver. The obtained results suggest that GVNS is able to find high-quality solutions in short amount of time.
This is a joint work with Luka Matijević and Tatjana Davidović presented at SYMOPIS 2021.
Utorak, 23.11.2021. u 14:15, Online
Predrag Veličković, Fakultet organizacionih nauka
MODEL INTERAKCIJE ČOVEKA I RAČUNARA KAO OSNOVA ZA UMANJENJE NEGATIVNIH EFEKATA APLIKACIJA SA VIRTUELNOM REALNOŠĆU NA PSIHOFIZIČKE KARAKTERISTIKE KORISNIKA
Tehnologija sistema virtuelne realnosti beleži ubrzani razvoj i sve veće interesovanje poslednje decenije, ali ujedno su uočeni negativni simptomi i efekti koji se javljaju kod korisnika pri interakciji sa virtuelnim okruženjem. Ranija istraživanja imala su fokus na razvoju i primeni same tehnologije, a tek poslednjih godina se akcenat stavlja na probleme pri korišćenju koji uključuju i ljudski faktor. Primarni cilj istraživanja je kreiranje modela interakcije čoveka i računara kao osnove za umanjenje negativnih efekata aplikacija sa virtuelnom realnošću na psihofizičke karakteristike korisnika radi poboljšanja korisničkog iskustva. Umanjenje problema sajber bolesti može doprineti lakšem i bezbednijem korišćenju virtuelnih aplikacija što doprinosi unapređenju tehnologije virtuelne realnosti.
Utorak, 30.11.2021. u 14:15, Online
Bogdana Stanojević, Matematički institut SANU
ON FINDING WELL SPREAD NON-DOMINATED VECTORS TO MANY-OBJECTIVE NONLINEAR OPTIMIZATION PROBLEMS
In this talk a recent approach to generate well spread non-dominated vectors to a many-objective nonlinear optimization problem will be described. The approach employs an optimization model designed to yield certain non-dominated vectors that can fill gaps between already generated non-dominated vectors, providing a way to deal with the adjacency of generated non-dominated vectors and to quantify the gaps between them. A pseudo-randomized variant of the exact approach will be also presented. Ten problems from the literature, including bi-objective, 3-objective, 5-objective and 10-objective test instances with non-convex, disconnected or continuous Pareto frontier were selected for illustration. The inverted generational distance (IGD) and the hyper-volume (HV) were used as performance metrics to measure the quality of the obtained approximations.
The presentation is based on a joint work with prof. Fred Glover, University of Colorado.
RUKOVODIOCI SEMINARA
MI SANU
Vera Kovačević-Vujčić
Milan Dražić
FON
Zorica Bogdanovic
Marijana Despotovic-Zrakic
IEEE
Bozidar Radenkovic
