magistrsko delo
Abstract
Magistrsko delo ima namen preizkusiti metodo prenosnega učenja na obdelavi naravnega jezika in jo primerjati s klasičnimi metodami učenja nevronskih mrež, metodo LSTM. V delu sta uporabljena opisna metoda za teoretični in eksperiment za praktični del dela. V slednjem smo ugotovili, da je metoda prenosnega učenja na majhni količini podatkov bolj točna od klasičnih metod, vendar za to potrebuje več časa. Delo primerja prednaučeni model Bert in klasično metodo LSTM, zato je priporočljivo primerjati rezultate tudi z drugimi prednaučenimi modeli in klasičnimi metodami.
Keywords
nevronske mreže;prenosno učenje;klasifikacija besedila;magistrske naloge;
Data
Language: |
Slovenian |
Year of publishing: |
2020 |
Typology: |
2.09 - Master's Thesis |
Organization: |
UM FERI - Faculty of Electrical Engineering and Computer Science |
Publisher: |
[J. Žerak] |
UDC: |
004.85(043.2) |
COBISS: |
44103683
|
Views: |
362 |
Downloads: |
72 |
Average score: |
0 (0 votes) |
Metadata: |
|
Other data
Secondary language: |
English |
Secondary title: |
Classification of text using transfer learning |
Secondary abstract: |
The aim of this Master's thesis is to test the method of transfer learning with natural language processing and to compare it to a standard neural network model, namely LSTM. The thesis is using the descriptive method for the theoretical part and experimental method for the practical part. In the experiment we have discovered that, while transfer learning is more accurate than the standard model, it is also slower in the learning process. The thesis compares only the pretrained model Bert and standard model LSTM and that is why it is recommended to also check other pretrained models and standard models for comparison. |
Secondary keywords: |
neural networks;transfer learning;NLP;PyTorch;LSTM; |
Type (COBISS): |
Master's thesis/paper |
Thesis comment: |
Univ. v Mariboru, Fak. za elektrotehniko, računalništvo in informatiko, Informatika in tehnologije komuniciranja |
Pages: |
IX, 64 f. |
ID: |
12104482 |