diplomsko delo
Jan Kuhta (Author), Zoran Bosnić (Mentor)

Abstract

V diplomskem delu smo združili področji računalniškega generiranja slik ter inkrementalnega učenja s poudarkom na zaznavanju spremembe učnega koncepta. Razvili smo inovativno metodo za zaznavanje sprememb v slikovnih podatkovnih tokovih, ki temelji na analizi izgub diskriminatorja znotraj zanke učenja generativnih nasprotniških mrež (GAN). Eksperimenti so pokazali, da se za naše namene najbolje obnese arhitektura Wasserstein GAN z gradientno kaznijo (WGAN-GP), ki smo jo uporabili v glavnem delu našega testiranja. Zasledili smo, da se ob ne\-nad\-nih spremembah porazdelitve vhodnih podatkov izguba diskriminatorja za določeno število iteracij strmo poveča po absolutni vrednosti. To specifično lastnost smo izkoristili pri razvoju metode Gan Loss Drift Detection (GLDD). Metodo smo temeljito testirali na slikovni množici MNIST digits, ki smo jo predhodno preoblikovali v deset različnih porazdelitev podatkov. Med poskusi se je posebej uspešna izkazala verzija GLDD-KSWIN, ki je dosegla povprečno natančnost 0.79, občutljivost 0.90, F1-oceno 0.84 in zamik zaznave 41.64. Rezultati kažejo, da je predlagana metoda obetaven temelj za nadaljnje raziskave na tem področju, ki še vedno ostaja precej nedotaknjeno in polno izzivov.

Keywords

strojno učenje;sprememba koncepta;generativne nasprotniške mreže;inkrementalno učenje;univerzitetni študij;diplomske naloge;

Data

Language: Slovenian
Year of publishing:
Typology: 2.11 - Undergraduate Thesis
Organization: UL FRI - Faculty of Computer and Information Science
Publisher: [J. Kuhta]
UDC: 004.85(043.2)
COBISS: 188026883 Link will open in a new window
Views: 27
Downloads: 5
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: Concept drift detection in incremental learning of generative adversarial networks
Secondary abstract: In this thesis, we combined the fields of computational image generation and incremental learning, with a focus on concept drift detection. We developed an innovative method for detecting conceptual changes in image data streams based on discriminator loss analysis within a generative adversarial network (GAN) learning loop. We conducted research that showed that the Wasserstein GAN architecture with gradient penalty (WGAN-GP), which we used in the main part of our testing, performed best for our purposes. We observed that when the distribution of the input data changes abruptly, the loss of the discriminator for a given number of iterations increases sharply in absolute value. We exploited this specific property in the development of the Gan Loss Drift Detection (GLDD) method. We thoroughly tested the method on an image dataset MNIST digits, which we had previously transformed into ten different distributions. During the experiments, the GLDD-KSWIN version performed particularly well, achieving an average precision of 0.79, a recall of 0.90, an F1-score of 0.84, and a latency of 41.64. The results show that the proposed method provides a promising foundation for further research in this area, which still remains largely untouched and challenging.
Secondary keywords: machine learning;concept drift;deep learning;generative adversarial networks;incremental learning;computer and information science;diploma;Globoko učenje (strojno učenje);Računalništvo;Univerzitetna in visokošolska dela;
Type (COBISS): Bachelor thesis/paper
Study programme: 1000468
Embargo end date (OpenAIRE): 1970-01-01
Thesis comment: Univ. v Ljubljani, Fak. za računalništvo in informatiko
Pages: 47 str.
ID: 23073801