Predavanja možete pratiti i online putem MITEAM stranice Seminara za računarstvo i primenjenu matematiku:
https://miteam.mi.sanu.ac.rs/asset/qGapAHyEBad2FDwXR
PLAN RADA SEMINARA ZA MAJ 2026. GODINE
Utorak, 12.05.2026. u 14:15, Knez Mihailova 36, sala 301f i
Online
Veljko Jeremić i Veljko Uskoković, Fakultet organizacionih nauka, Univerzitet u Beogradu
A DIGITAL TWIN FOR PLATFORM GOVERNANCE: SIMULATING COLD-START TRADEOFFS WITH LLM-POWERED AGENTS
Generative and predictive AI are automating many remote gig tasks, yet location-based gig work remains inherently physical and continues to expand into trust-intensive home services. In these markets, new platforms face a cold-start problem in which early growth stalls because users and providers must decide whether to join before they can observe reliable quality, safety, and governance. Although the literature has examined individual governance mechanisms (e.g. referral incentives, reputation systems, provider certification, loyalty programmes), their joint effects on market formation remain under-researched because these mechanisms are typically studied in isolation. We address this gap through a two-phase research design grounded in signalling theory and mechanism design. Phase 1 maps actor-specific trust antecedents through platform orchestrator interviews and surveys of users and providers, combining descriptive insights with structural equation modelling and choice-based conjoint analysis. Those results calibrate Phase 2, which deploys an empirically grounded generative agent-based model in which 1,000 heterogeneous LLM-powered agents make autonomous decisions across nine governance configurations, each executed for 30 independent replications (270 simulation runs, 1,709,079 LLM-generated decisions). A multi-objective Pareto dominance analysis across six performance dimensions identifies five non-dominated governance configurations on a governance frontier, revealing that no single mechanism portfolio optimises all objectives simultaneously. We contribute a governance-frontier approach to multi-objective platform design and advance methodological practice for LLM-powered simulation.
Sreda, 13.05.2026. u 19:00, Online
Tatjana Jakšić Krüger, Matematički institut SANU
PARALLEL VARIABLE NEIGHBORHOOD SEARCH: FROM STRUCTURED SEARCH TO MULTI-AGENT SYSTEMS
In an era dominated by data-driven approaches, it is often assumed that effective search requires learning from data. We revisit Variable Neighborhood Search (VNS), a classical trajectory-based metaheuristic and argue that its performance relies on a combination of problem-informed design and systematic neighborhood exploration guided by a fixed neighborhood-change rule. On one hand, VNS incorporates structural knowledge through the definition of neighborhood systems and solution representations, reflecting the expertise of the algorithm designer. On the other hand, during execution, the neighborhood-change rule systematically redirects the search trajectory in response to local search outcomes -- without modifying the rule itself. Focusing on parallel VNS, we adopt the established High-Level Taxonomy of Parallel Metaheuristics (HTPM) and the complementary Implementation Taxonomy (ITPM) to analyze different parallelization strategies. These frameworks characterize algorithms along dimensions such as decomposition (independent vs. cooperative search), control (centralized vs. decentralized), communication (synchronous, asynchronous, or knowledge-based), solution diversity, granularity, and computing architecture. Within this perspective, parallel VNS is not merely a speed-up of the sequential algorithm, but a family of multi-search methods whose behavior depends fundamentally on the selected HTPM/ITPM configuration. Independent and cooperative strategies enable qualitatively different modes of parallel search, affecting both the exploration-exploitation balance and the robustness of the overall method. We show that these taxonomic dimensions naturally correspond to established AI concepts -- including inductive bias, exploration-exploitation tradeoff, and multi-agent coordination -- providing a unified perspective on parallel VNS as an AI method.
Zajednički sastanak sa seminarom Veštačka inteligencija.
Utorak, 19.05.2026. u 14:15, Knez Mihailova 36, sala 301f i
Online
Nikola Zogović, Institut „Mihajlo Pupin“
VIŠECILJNA OPTIMIZACIJA U INŽENJERSTVU
Edžvort i Pareto su dali prve formulacije ideje višekriterijumske i višeciljne optimizacije još krajem 19. i početkom 20. veka. Međutim, značajniji razvoj i primena su sačekali razvoj računarske tehnologije sa kojom su u tesnoj korelaciji. Ubrzanje primene je takvo, da je skoro šestina objavljenih dokumenata iz poslednjih skoro 60 godina, na temu višeciljne/višekriterijumske optimizacije u Skopus bazi, iz 2024. godine. U ovom predavanju će biti izloženi izazovi i trendovi u višeciljnoj optimizaciji sistema. Biće prikazane osnove šestofazne metodologije koju autor koristi i rezvija u svom istraživačkom radu u Institutu „Muhajlo Pupin“, kao i više primera primene u aktuelnim istraživačkim oblastima, kao što su internet stvari i bežične senzorske mreže, radarska tehinka, pametni gradovi, kibernetsko-fizičko pčelarstvo, Industrija 4.0 ili fabrike budućnosti, računarstvo u oblaku i bioinformatika. Na kraju će biti diskutovane otvorene teme kao što su višeciljne optimizacija i pronalazaštvo, redukcija višeciljnog problema na bitan problem, višeciljna optimizacija i održivost i pitanje konvergencije skupa optimalnih rešenja sa porastom broja ciljeva.
Utorak, 26.05.2026. u 14:15, Knez Mihailova 36, sala 301f i
Online
Božidar Radenković, FON, Univerzitet u Beogradu
OSNOVNI IZVRŠNI SUPSTRAT (ZTES) ZA EVOLUTIVNE SOFTVERSKE SISTEME
Savremeni softverski i AI sistemi suočavaju se sa nedostatkom jedinstvenog formalnog rešenja za probleme determinističke reproduktivnosti, konzistentnosti stanja u realnom vremenu i ekvivalentnosti upravljačkih i operativnih procesa. Uzrok je ontološke prirode: ne postoji formalno definisan supstrat u kojem bi izvršavanje, perzistencija i AI rezonovanje delili jedinstvenu, kauzalno uređenu strukturu znanja. Ovaj rad predstavlja Zero Tier Execution Substrate (ZTES) — aksiomatski model nastao formalnom sintezom Mesarović–Takahara ontologije sistema, Lamportovog kauzalnog poretka i DEVS formalizma. Kroz trofazno jezgro izvršavanja, istorijska baza podataka koja se isključivo dopunjuje (append-only) postaje primarni računarski medijum. U njemu su upravljanje i operacije formalno ekvivalentni procesi tranzicije, čime se izvršavanje sistema poistovećuje sa kauzalnom evolucijom znanja: Execution(Σ) ≡ Evolution(K). Model je univerzalan za diskretne procese i ne uvodi nove primitive, već definiše strogu semantičku disciplinu nad postojećim mehanizmima.
ZTES osigurava determinističku serijalizaciju događaja i korekciju stanja bez destruktivnih izmena, dok se Šesta normalna forma (6NF) pojavljuje kao prirodna ontološka posledica atomskih događaja. Ovakva arhitektura direktno omogućava sledljivost AI odlučivanja i strukturno nametnutu autonomiju. Time se softverski i AI sistemi redefinišu: oni više nisu slojevite arhitekture, već kauzalno evoluirajuće strukture znanja vođene precizno definisanom semantikom izvršavanja. Tekst kompletnog rada se može preuzeti sa adrese:
https://doi.org/10.5281/zenodo.19183067.
Četvrtak, 27.05.2026. u 12:00, Online
Raka Jovanović, Qatar Environment and Energy Research Institute
BEYOND THE BLACK BOX: BRIDGING THEORETICAL PROMISE AND PRACTICAL REALITY IN APPLIED NEURAL NETWORKS
This lecture bridges the gap between the theoretical promise of neural networks and their practical performance in real-world forecasting tasks. Drawing on case studies in air quality forecasting, graph combinatorial optimization, crop import price prediction, food delivery demand forecasting, and wildfire spread modeling, we show that architectural complexity alone is rarely the key to success. Across these domains, simple feature engineering—such as value-adaptive smoothing, rolling statistics, and lagged variables—often improves accuracy more than switching from LightGBM to LSTMs or Transformers. Tree-based and robust regression models consistently match or outperform deep networks, especially when data is noisy, sparse, or contains outliers. Moreover, standard loss functions frequently misalign with operational goals, making asymmetric or robust losses essential for practical deployment. The talk offers a pragmatic framework for choosing models based on data constraints, computational resources, and domain-specific objectives, rather than theoretical elegance alone.
Zajednički sastanak sa seminarom Veštačka inteligencija.
RUKOVODIOCI SEMINARA
MI SANU
Vera Kovačević-Vujčić
Milan Dražić
FON
Zorica Bogdanovic
Marijana Despotovic-Zrakic
IEEE
Bozidar Radenkovic
