diplomsko delo
Timotej Košir (Author), Jure Žabkar (Mentor)

Abstract

Nevronske mreže na področju strojnega učenja, predvsem pri obdelavi slik, besedil in zaporedij, prekašajo klasične modele. Njihova popularnost je prebudila poskuse iskanja alternativ. Kot ena izmed idej so se pojavili globoki gozdovi, ki jih avtorja vpeljeta v obliki strukture gcForest. Temelji na modelu naključnih gozdov, ki jih združujemo v kaskade. Struktura gcForest se odlikuje na učenju z majhnimi podatkovnimi množicami in ne potrebuje veliko računalniških virov. Raziščemo osnovne gradnike strukture in podamo spremembe, ki bi lahko izboljšale delovanje. Temu dodamo lastno implementacijo dreves, naključnih gozdov ter strukture s predlaganimi spremembami. Posamezne dele implementacije testiramo. Bolj se posvetimo primerjavi rezultatov z različico avtorjev, ki jo opravimo na štirih podatkovnih množicah. Ker dosežemo rahlo slabše rezultate, podamo razloge zanje. Govorimo tudi o naslednjih korakih implementacije.

Keywords

gcForest;naključni gozdovi;zlaganje;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: [T. Košir]
UDC: 004.8(043.2)
COBISS: 123848707 Link will open in a new window
Views: 33
Downloads: 8
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: Implementation of deep forest structure based on gcForest
Secondary abstract: In the field of machine learning, especially in image, text, and sequence processing, neural networks have had more success than classical models. The popularity of neural networks has led to several attempts to find an alternative. Deep forest structures have emerged as one of them. The authors have integrated the concept of deep forest into a model called gcForest. It is based on robust random forests that are connected into a cascade. Structure excels at learning from small data sets and does not consume many computer resources. We examine the building blocks of the structure and introduce changes that could improve performance. In the next step, we program a custom library for decision trees, random forests, and the structure with the proposed changes. Then we test the parts of the implementation separately and compare the results with those of the authors on four data sets. Since we obtain worse results, we provide an explanation. Finally, we talk about the next steps of the implementation.
Secondary keywords: gcForest;random forests;deep learning;stacking;computer 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: 39 str.
ID: 16479270