Seminar for
DECISION MAKING – THEORY, TECHNOLOGY AND PRACTICE

 

PROGRAM

---

Predavanja možete pratiti i online putem MITEAM stranice Seminara Odlučivanje - teorija, tehnologija, praksa:
https://miteam.mi.sanu.ac.rs/asset/sEL32w8mjmruyeEqW

---

Plan rada Seminara Odlučivanje - teorija, tehnologija, praksa za OKTOBAR 2026.

---

Četvrtak, 08.10.2026. u 13:00, Online
Sandra Stanković, University of Nis, Faculty of Electronic Engineering Niš
ADAPTIVE TIME–FREQUENCY ANALYSIS AND EFFICIENT DEEP LEARNING FOR BIOMEDICAL SIGNAL PROCESSING
Real world signals are complex, non-stationary signals whose relevant information is often distributed across multiple temporal and spectral scales. While deep neural networks have demonstrated strong capabilities for direct end-to-end analysis of 1D signals, their computational demands, memory requirements, and limited interpretability can restrict their deployment in wearable and resource-constrained systems. Combining advanced signal representations with lightweight neural network architectures provides a promising approach for developing efficient, accurate, and reliable intelligent sensing solutions. This presentation explores the role of time–frequency analysis in modern biomedical signal processing, focusing on how carefully designed representations can reduce the computational complexity of learning-based models while preserving important physiological information. Classical approaches, including Fourier analysis, discrete cosine transform (DCT), short-time Fourier transform (STFT), and wavelet-based representations, will be introduced, together with their advantages and limitations for non-stationary biomedical signals. The presentation will further discuss adaptive and learnable time-frequency methods which enable task-dependent optimization of the representation. In addition, lightweight neural network design strategies, including efficient convolutional architectures and compact models suitable for edge deployment, will be presented as a complementary approach to reducing computational cost. Applications from physiological signal analysis, including ECG, PPG, and non-invasive monitoring, will be used to demonstrate how the integration of signal processing, adaptive representations, and efficient deep learning architectures can enable accurate and computationally sustainable biomedical AI systems.

---

Četvrtak, 15.10.2026. u 13:00, Pariske Komune bb, Niš i Online
Katica (Stevanović) Hedrih, Mathematical Institute of the Serbian Academy of Sciences and Arts
SECOND FRACTIONAL ORDER DIFFERENTIAL CONSTITUTIVE RELATIONS FOR VISCOELASTIC MATERIALS OF THE FRACTIONAL TYPE WITH PIEZOELECTRIC PROPERTIES, REPRESENTING RHEOLOGICAL MODELS OF HIGHER COMPLEXITY LEVELS
The lecturer presents newly derived differential constitutive relations (equations) of the second fractional order for complex, higher-level rheological models of fractional-type viscoelastic materials with piezoelectric properties. These newly derived differential equations are of the second fractional order and incorporate Caputo differential operators of the first and second fractional orders. The lecturer outlines an approximate analytical methodology for solving these second-fractional-order differential equations, employing the Laplace transform, expansion into power series with respect to the complex Laplace parameter, and the properties of the convolution of three functions. This methodology yields approximate analytical solutions for the mew derived second-fractional-order differential constitutive equations governing these complex, higher-level rheological materials. These approximate analytical solutions simultaneously serve as integral constitutive relations, establishing the link between normal stresses and axial strains in these complex, higher-level rheological models of fractional-type viscoelastic materials with piezoelectric properties. The theoretical analytical results obtained pertain to complex fractional-type material models, specifically the second-level complexity Letherish-Faraday model for a fractional-type viscoelastic material with piezoelectric properties; The Jeffreys-Faraday model represents a second-level complexity rheological model for a fractional-type viscoelastic material with piezoelectric properties, while the Burgers-Faraday model represents a third-level complexity model of the same type. These three rheological models of higher complexity—in their classical linear forms—are already widely used; the author believes that the newly introduced fractional-type models (named by the author) will also see widespread application, initially in the validation of experimental research and subsequently in practical use. The lecturer also believes that these results open up new challenges regarding the development of software for calculating inverse Laplace transforms of complex Laplace-transformed expressions, thereby expanding the scope of artificial intelligence in areas where it is currently underdeveloped and limited

---

---