Method for linear layout generation of correlated elements using artificial intelligence

[J. Lj. Pejić]

119 str.

Biografija: str. 117-119. Bibliografija: str. 101-111.

Artificial intelligence

The aim of this dissertation is to formally define and address the problem of linear arrangement of correlated elements, which appears across numerous domains. As a solution, a method based on convolutional neural networks is proposed, utilizing an innovative data encoding approach through multichannel binary sequences. This approach enables efficient learning of spatial relationships and precise dimensional reasoning, even under conditions of limited data availability. A dataset was developed for evaluating the spatial reasoning capabilities of machine learning models. Experimental analyses demonstrated that the proposed method successfully learns fundamental spatial relations and outperforms existing approaches, particularly in scenarios with a small number of training examples. The method was applied in two domains: architecture, where it was used for automatic generation of linear kitchen layouts, and beekeeping, where it was applied to the problem of predicting hive weight changes based on meteorological data. In both cases, its applicability and robustness were confirmed. The dissertation thus contributes to the formalization of an important practical problem, the development of a generic method for its solution, the creation of evaluation datasets, and the demonstration of its broad applicability in real-world environments.

veštačka inteligencija, konvolucione neuronske mreže, prostorno rezonovanje, dizajniranje rasporeda, generisanje enterijera, precizno pčelarstvo, predikcija težine košnice

artificial intelligence, convolutional neural networks, spatial reasoning, layout design, interior design, precision beekeeping, hive weight prediction

519.852:004.8(043.3)

P 176

Tekst