MISANU

PROGRAM

Predavanja možete pratiti i online putem MITEAM stranice Seminara iz veštačke inteligencije:
https://miteam.mi.sanu.ac.rs/asset/DPP9i2jhvYzp8dmRe

Plan rada Seminara iz veštačke inteligencije za JUN 2026.

Utorak, 02.06.2026. u 14:15, Knez Mihailova 36, sala 301f i Online
Veljko Milutinović, Elektrotehnički fakultet, Univerzitet u Beogradu
ENTREPRENEURSHIP
This presentation describes the course on the topic from the title.
This topic can be covered as an isolated lecture of one to four academic hours or as a full one-semester course of 12 to 36 academic hours with 12 homework assignments. It covers 12 management-oriented high-tech supported paradigms for creative and holistic management of projects in any field of interest, with the stress on AI and Big Data.
Themes included: (a) Writing proposals for the NSF in the USA or for Horizon in Europe, (b) Mastering the elevator-pitch generation skills, by writing short analytical esseys for GMAT or GRE, (c) Strategic planning using CMMI of DARPA, (d) Tactical planning using Agile Methods of DARPA, (e) Creating a Highly Visible web presence for Worldwide Markets, a strategy created at Harvard, (f) Creating the Mind Genomics Campaign for Targeted Marketing, a strategy created at Harvard, (g) Incorporating based on SBA of USA, (h) Protecting based on PTO of USA, (i) Writing survey articles for SCI journals, (j) Writing research articles for SCI journals, (k) Data Mining and Semantic Web, and (l) Creativity and Branding.
Each theme includes theory, practice, examples, and an anectode that stresses the essence. So far, in the last 40 year, the course was presented, in its various forms, for years in the line, at Indiana University, Purdue University, MIT, and Harvard in the USA. In Asia at Hebrew University in Jerusalem, Technion, Bogazici, Koç, Thinghua, Shandong, Sendai, and Tokyo. Also, in Europe at ETH, EPFL, UNIWIE, TUWIEN, Siena, Salerno, Barcelona, Madrid. In exYu at Belgrade, Podgorica, Zagreb, and Ljubljana.
The presentation is based on author's experiences from his DARPA projects on GaAs and DataFlow computing, and is inspired by current VicePresidents (for the company's core business) of Qualcomm, Intel, IBM, AMD, NCR, and HP Labs, who all graduated from the University of Belgrade, where this course is/was obligatory or elective in four different schools (ETF, MATF, GRF, FON).
In the current academic year, or in the very near past, the course has been taught also for students of Purdue University, Indiana University in Bloomington, MIT, Harvard, Hebrew University of Jerusalem, Tel Aviv University, Technion, Weizmann, Universities of Salerno and Siena, as well as of UNIWIE and TUWIEN.
The course is supported with a textbook by Cambridge Publishers, for which the pearls of wisdom were contributed by 20 Nobel Laureates.
Zajednički sastanak sa Seminarom za računarstvo i primenjenu matematiku.

Sreda, 03.06.2026. u 19:00, Online
Lluís Godo Lacasa, Artificial Intelligence Research Institute (IIIA) of the Spanish National Research Council (CSIC)
TWO APPROACHES TO BOOLEAN CONDITIONALS WITHIN BOTH THE PROBABILISTIC AND POSSIBILITIC FRAMEWORKS
Conditionals play a key role in different areas of logic, probabilistic and non-monotonic reasoning, and they have been studied and formalised from different angles. In this talk we will focus on recent developments on various foundational aspects of conditionals related to the probabilistic and possibilistic models of uncertainty. First we will show that any plain probability on an algebra of events can be extended to a bigger Boolean algebra of (compound) conditionals in such a way that the (plain) probability of a conditional is its conditional probability. Hence a version of the Stalnaker's Thesis holds in this setting. Then we will also show that a suitable notion of conditional possibility, for which a triviality result similar to Lewis' in the probabilistic setting can be proved, is fully compatible with the above algebraic setting of Boolean algebras of conditionals in the sense that an analogue of Stalnaker's Thesis holds as well. On the other hand, we will consider a different approach to conditionals as random quantities by Gilio, Sanfilippo and colleagues, based on de Finetti's notion of conditional as a three-valued object, and we will show that it is not only compatible with the above Boolean algebraic setting, but it also admits a faithful possibilistic counterpart. In the latter case, conditionals are interpreted as possibilistic variables instead, and their possibilistic expectation (a Sugeno-like integral) provides a means of extending a possibility on plain events to arbitrary (compound) conditionals.

Sreda, 10.06.2026. u 19:00, Online
Vesna Marinković, Matematički fakultet, Univerzitet u Beogradu
INTELIGENTNI SISTEMI ZA REŠAVANJE GEOMETRIJSKIH KONSTRUKCIJA I DOKAZIVANJE TEOREMA
Još od samih početaka razvoja veštačke inteligencije, geometrija je predstavljala interesantan domen istraživanja. Tako je Herbert Gelernter još 1959. godine razvio sistem za automatsko dokazivanje geometrijskih teorema, sposoban da uspešno dokazuje mnoge teoreme euklidske geometrije. Ipak, geometrija i danas ostaje izazov za alate veštačke inteligencije zbog čega se razvijaju različiti sistemi koji koriste tehnike mašinskog učenja, poput sistema AlphaGeometry. Naime, pokazuje se da geometrija zbog svojih specifičnosti predstavlja poseban izazov za automatizaciju. Upravo zato pogodna je za ispitivanje sposobnosti AI alata da vrše logičko zaključivanje i da izvode dokaze, ali i za postavljanje pitanja objašnjivosti - razumevanja zbog čega se neki dokaz smatra ispravnim.
U ovom predavanju biće reči o alatima koji omogućavaju automatsko rezonovanje u geometriji, kao i o izazovima koji ovu oblast i danas čine relevantnom. Fokus predavanja biće na problemu rešavanja konstruktivnih geometrijskih zadataka. Biće prikazan automatski sistem za njihovo rešavanje, kao i način da dokažemo da konstruktivni problem nije rešiv. Jedan od ključnih aspekata rešenja konstruktivnog zadatka jeste dokazivanje korektnosti generisanih rešenja: disktuvaćemo o tome da li je moguće dobiti čitljive dokaze, koje i čovek može da razume. Pričaćemo i o mogućnosti da se ovaj sistem proširi kako bi mogao da vrši navođenje studenata tokom procesa rešavanja.

Sreda, 17.06.2026. u 19:00, Online
Vilém Novák, University of Ostrava, Institute for Research and Applications of Fuzzy Modeling, Ostrava, Czech Republic
INTERMEDIATE QUANTIFIERS AND THEIR SYLLOGISMS IN FUZZY NATURAL LOGIC AND AI
Artificial Intelligence includes many theories and methods that are capable of performing tasks associated with human intelligence. Among them, the leading role is played by techniques of machine learning and neural networks. We argue that an important role in AI also plays formal logic. In this talk we will mention the concept of Fuzzy Natural Logic (FNL), which is a system of theories of higher-order mathematical fuzzy logic enabling us to model special ways of human thinking that is based on the use of natural language. FNL stems from the results of classical linguistics, logical analysis of concepts and semantics of natural language and is formalized using higher-order mathematical fuzzy logic.
A constituent of FNL is the theory of fuzzy generalized quantifiers that are special natural language expressions using which we quantify the number of elements in various contexts. Typical examples are "Many, most, a lot of, a few, several, almost all" that form a subclass called "intermediate quantifiers". In this talk we will present results obtained in a formal theory of the latter. We will also describe reasoning using generalized syllogisms in which intermediate quantifiers occur. We argue that methods of formal logic are indispensable for their theory because using the latter we can distinguish valid syllogisms from those which are not valid and, therefore, cannot be used in human reasoning. We will also show that AI without logic fails when checking validity of syllogisms. We will demonstrate valid as well as invalid syllogisms on examples. Along with it, we will discuss graded square of opposition as a general scheme for human reasoning.
Finally we will touch non-monotonic reasoning and show that our theory is capable of solving typical problems of it. Namely, we will prove that the classical examples "Most birds fly" and "Tweety is a penguin which does not fly" do not lead to contradiction in our theory. We argue that this is the consequence of the capability of FNL to model the meaning of vague concepts.

Sreda, 24.06.2026. u 19:00, Online
Biljana Stojanović, Matematički institut SANU
Nenad Mitić, Matematički fakultet, Univerzitet u Beogradu
TEMPORAL DYNAMICS OF CODON USAGE IN SARS-CoV-2 PROTEINS
Codon usage bias (CUB) reflects the non-random usage of synonymous codons and represents an important feature associated with viral evolution, host adaptation, and translational efficiency. We analysed CUB in approximately 5 million SARS-CoV-2 protein-coding sequences collected between 15 December 2019 and 22 May 2023. The analysis was performed on the complete dataset of protein amino acid sequences and corresponding coding sequences, as well as on 20 temporal subsets obtained by partitioning isolate collection dates into discrete time intervals. Four proteins — surface glycoprotein, nucleocapsid phosphoprotein, ORF1a polyprotein, and ORF1ab polyprotein — were selected for detailed analysis, each exhibiting more than 3% unique coding sequences relative to the total number of sequences.
For each protein type, we analyzed (i) codon usage patterns across all corresponding coding sequences using global codon frequencies and Relative Synonymous Codon Usage (RSCU) values as codon-specific CUB measures, (ii) data derived from alignments of individual amino acid sequences against the corresponding amino acid sequence of the reference isolate, and (iii) coding sequences corresponding to aligned amino acid sequences.
Low-frequency codons were identified across the analyzed datasets. Specific positions within each reference protein at which significant changes in codon frequencies were observed over time were identified. The analysis also examined codon usage patterns in relation to the WHO-defined lineage associated with each isolate. In addition, CUB was analyzed for each protein type at the level of complete coding sequences using the gene-specific measures Effective Number of Codons (ENC) and Relative Codon Bias Score (RCBS). These measures provided an objective characterization of CUB for each analyzed protein type and complemented the codon-specific analysis based on global codon frequencies and RSCU values.
The results indicate that the most pronounced changes in codon usage over time were observed in surface glycoprotein and nucleocapsid phosphoprotein. Some synonymous codons show relatively stable frequencies over time, whereas others exhibit decreases in codon abundance at different stages (early, intermediate, or late) and with varying magnitudes.

Utorak, 30.06.2026. u 14:15, Knez Mihailova 36, sala 301f i Online
Nataša Pržulj, Mohamed bin Zayed University of Artificial Intelligence, Abu Dhabi, UAE; Department of Computer Science, University College London, London, United Kingdom
MULTI-OMICS DATA FUSION FOR PRECISION MEDICINE AND THERAPEUTICS
Large amounts of multi-omic data are increasingly becoming available. They provide complementary information about cells, tissues and diseases. We need to utilize them to better stratify patients into risk groups, discover new biomarkers and targets, re-purpose known and discover new drugs to personalize medical treatment. This is nontrivial, because of computational intractability of many underlying problems on large interconnected data (networks, or graphs), necessitating the development of new algorithms for finding approximate solutions (heuristics).
We develop versatile artificial intelligence (AI) frameworks for multi-omics data fusion, constrained by the state-of-the-art network science methods, to address key challenges in precision medicine and pharmacology from time-series, multi-omics data, including patient-derived single-cell data, to: better stratify patients, predict new biomarkers and targets, re-purpose existing and discover new drugs; we apply these to different types of cancer, Covid-19, Parkinson’s and other diseases. Our new methods stem from graph-regularized non-negative matrix tri-factorization (NMTF), a machine learning (ML) technique for dimensionality reduction, inference, fusion and co-clustering of heterogeneous datasets, coupled with novel graphlet-based network science algorithms. We utilize our new frameworks for improving the understanding of the molecular organization of life and of diseases from the embedding spaces of omics data. Also, we utilize the local network topology to correct for the topological information missed by random walks used in many ML methods, and to enable embedding of multi-omics networks into more linearly separable spaces, allowing for their explainable and sustainable mining. The aim is to develop an overreaching framework encompassing all multi-omics data towards consumer-facing precision medicine products
Zajednički sastanak sa Seminarom za računarstvo i primenjenu matematiku.

Ovaj onlajn seminar nastao je kao nastavak sastanka "Serbian AI Meeting" i zamišljen je da na njemu istraživači iz Srbije i iz dijaspore, kao i istraživači sa univerzteta, naučnih instituta i iz prakse predstavljaju naučne teme i rezultate iz oblasti veštačke inteligencije.
Link za svako pojedinačno predavanje biće dostavljen dan pre održavanja predavanja.

Andreja Tepavčević
Rukovodilac seminara

Biljana Stojanović
Sekretar seminara