PLAN RADA SEMINARA ZA JUN 2022. GODINE
Predavanja na seminaru mogu se pratiti na daljinu 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, 07.06.2022. u 14:15, Knez Mihailova 36, sala 301f i
Online
Emil Jovanov, Electrical and Computer Engineering Department at the University of Alabama in Huntsville
SEAMLESS HEALTH MONITORING: OPPORTUNITIES AND CHALLENGES
Ubiquitous smart sensors and devices integrated into the Internet
of things (IoT) environment transformed our homes, offices, and industries,
while wearable monitoring and mobile health (mHealth) revolutionized
healthcare diagnostics and delivery. Smart environments create unprecedented
opportunities for seamless health monitoring and preventive diagnostics.
“Things” with embedded activity and vital sign sensors that we refer to as
“smart stuff” can interact with wearable and ambient sensors in Synergistic
Personal Area Networks—SPANs. SPANs creates a dynamic “opportunistic bubble”
for ad-hoc integration with other sensors of interest around the user,
wherever they go. The synergy of information from multiple sensors can
provide: (a) New information that cannot be generated from existing data
alone, (b) user identification, (c) more robust assessment of physiological
signals, and (d) automatic annotation of events/records. This lecture
presents possible new applications, opportunities, and challenges of new
environments.
Utorak, 14.06.2022. u 14:15 Online
Ilir Čapuni, Prirodno-matematički fakultet Univerziteta Crne Gore
KONSTRUKCIJA TURINGOVE MAŠINE OTPORNE NA VJEROVATNOSNI ŠUM
Pouzdanost izračunavanja odnosi se na računanje pomoću mašine koja
je podvrgnuta određenom šumu. Nas prije svega interesuju prolazne greške,
tj. greške koje nijesu posljedica nekog kvara ili trajnog oštećenja neke
komponente, a javljaju se nezavisno jedan od drugog sa nekom malom
vjerovatnoćom.
Istorijski prvi rezultat ove vrste je rad von Neumanna koji za svaki
Booleovo kolo $C$ veličine $n$ konstruiše novo kolo $C'$ veličine
$O(n\log{n})$ koji sa velikom vjerovatnoćom rješava isti zadatak kao i $C$,
iako svaki gejt kola $C'$ može da pogreši sa nekom malom
vjerovatnoćom.
U ovom predavanju, daćemo konstrukciju univerzalne Turingovog mašine sa
jednom trakom koja može sprovesti proizvoljno dugačka izračunavanja
čak i uz prisustvo vjerovatnosnog šuma koji se definiše kako slijedi: u
svakom koraku, nezavisno od prethodnog koraka, promjena stanja glave,
aktivne ćelije i kretanje glave mogu, sa malom vjerovatnoćom, biti
različite od one koju diktira program mašine.
Konstrukcija je iznenađujuće složena i koristi hijerarhiju simulacija:
mašina $M_1$ simulira mašinu $M_2$ koji simulira mašinu $M_3$, i tako
dalje.
Mašina $M_1$ može “izdržati” šuma nivoa 1, $M_2$ može podnijeti šum 2.-og
nivoa, i tako dalje. Svaka od ovih mašina može se implementirati na
univerzalnoj mašini koristeći program $p$ i nivo $k$ kao ulazne podatke.
Program $p$ je zajednički program za sve ove mašine i on je ugrađen
na mašini $M_1$, pa se isti ne može oštetiti od grešaka. Forsiraćemo sve
nivoe da koriste isti program koristeći Kleen-ovu teoremu o fiksnoj
tački. Ovom konstrukcijom, izračunavanje koje traje $t$ koraka može se
simulirati u $t(\log{t})^{\alpha \log{\log{\log{t}}}}$ koraka uz
prisustvo vjerovatnosnog šuma, za neku konstantu $\alpha$.
Koautor ovog rada je Peter Gacs.
Utorak, 21.06.2022. u 14:15, Knez Mihailova 36, sala 301f i
Online
Luka Matijević, Matematički institut SANU
VARIABLE NEIGHBORHOOD SEARCH FOR MULTI-LABEL FEATURE SELECTION
With the growing dimensionality of the data in many real-world
applications, feature selection is becoming an increasingly important
preprocessing step in multi-label classification. Finding a smaller subset
of the most relevant features can significantly reduce resource consumption
of model training, and in some cases, it can even result in a model with
higher accuracy. Traditionally, feature selection has been done by employing
some statistical measure to determine the most influential features, but in
recent years, more and more metaheuristics have been proposed to tackle this
problem more effectively. In this lecture, we will present a brief
introduction to machine learning and data mining algorithms, with a focus on
feature selection. We will present different approaches to feature
selection, concentrating on metaheuristic wrapper methods. We will also
present the results of our recent work on this topic, where we proposed the
Basic Variable Neighborhood Search (BVNS) algorithm to search for the
optimal subset of features, combined with a local search method based on
mutual information. The algorithm can be considered a hybrid between the
wrapper and filter methods, as it uses statistical knowledge about features
to reduce the number of examined solutions during the local search. We
compared our approach against Ant Colony Optimization (ACO) and Memetic
Algorithm (MA), using the K-nearest neighbors classifier to evaluate
solutions. The experiments conducted using three different metrics on a
total of four benchmark datasets suggest that our approach outperforms ACO
and MA.
Utorak, 28.06.2022. u 14:15, Knez Mihailova 36, sala 301f i
Online
Jozo Dujmović, Department of Computer Science, San Francisco State University
STEPENOVANA LOGIKA (GRADED LOGIC)
Ljudske percepcije istinitosti, značaja, pogodnosti, preference, i mnoge
druge, imaju intenzitet u intervalu od najmanjeg (0) do najvećeg (1).
Te percepcije su stvar stepena, pa ih nazivamo stepenovane percepcije.
Stepenovana logika je kontinualna logika ljudskog rezonovanja sa
stepenovanim percepcijama. Cilj stepenovane logike je da ponudi matematičke modele
koji opisuju ljudsko rezonovanje u oblasti vrednovanja i odlučivanja na
način koji se može eksperimentalno verifikovati. Stepenovana logika je
konsistentna generalizacija klasične Bulove (G. Boole) logike. Bulova logika
funkcioniše u rogljevima jediničnog hiperkuba, a stepenovana logika
funkcioniše u celom jediničnom hiperkubu. Klasične logike obično potpuno
zanemaruju stepen značaja istinitosti; stepenovana logika ga eksplicitno
uzima u obzir. Prvi radovi iz stepenovane logike publikovani su u Beogradu
1973. godine. Šta se dešavalo sa ovom oblasti u narednih pola veka
vidi se u knjizi Jozo Dujmović,
Soft Computing Evaluation Logic, Wiley, 2018. Naš cilj je da ponudimo sažetu prezentacju ove oblasti.
RUKOVODIOCI SEMINARA
MI SANU
Vera Kovačević-Vujčić
Milan Dražić
FON
Zorica Bogdanovic
Marijana Despotovic-Zrakic
IEEE
Bozidar Radenkovic
