magistrsko delo
Abstract
Metode strojnega učenja vse bolj prodirajo v vsa področja modernega gospodarskega in raziskovalnega okolja. Obstoječi algoritmi dosegajo vrhunske rezultate pri nalogah, kot so prepoznavanje slik, razumevanje besedil in govora ipd. Avtomatizirane rešitve takšnih nalog so še nedavno veljale za nedosegljive. V tem magistrskem delu pregledamo najpopularnejše globoke nevronske mreže, iz njih sestavljene modele in njihove načine učenja. S pridobljenim znanjem in večkratnim testiranjem v drugem delu, razvijemo model globoke nevronske mreže za napovedovanje GPS sledi. Osnovno testiranje modela poteka na lastnem naboru sintetično ustvarjenih podatkov. Dva najuspešnejša modela v nadaljevanju učimo s pomočjo izbranih realnih podatkov, pridobljenih od podjetja GoOpti, d. o. o. Končni izpopolnjen model učimo z razširjenim naborom realnih podatkov. V magistrskem delu so opisani izbira in implementacija modela, način učenja, ustvarjanje in pridobivanje naborov podatkov in pridobljeni rezultati.
Keywords
magistrska dela;strojno učenje;globoko učenje;globoke nevronske mreže;povratne nevronske mreže;
Data
Language: |
Slovenian |
Year of publishing: |
2018 |
Typology: |
2.09 - Master's Thesis |
Organization: |
UM FNM - Faculty of Natural Sciences and Mathematics |
Publisher: |
[J. Borlinić] |
UDC: |
004.85(043.2) |
COBISS: |
24225288
|
Views: |
1337 |
Downloads: |
114 |
Average score: |
0 (0 votes) |
Metadata: |
|
Other data
Secondary language: |
English |
Secondary title: |
Predicting GPS tracks with deep neural networks |
Secondary abstract: |
Machine learning methods are increasingly influencing all areas of the modern economic and research environment. Existing algorithms achieve top results in tasks such as image recognition, understanding text and speech, etc. Automated solutions to such tasks were until recently considered unavailable. In this master's thesis, we review the most popular deep neural networks, underlying models and their learning tipes. With the acquired knowledge and repeated testing in the second part, we develop a deep neural network model for predicting GPS tracks. Basic testing of the model takes place on our own synthetically generated dataset. The two most successful models are further taught using selected real data obtained from GoOpti, d. o. o. and the final, best performing, model is taught with an expanded set of real data. The master's thesis describes the choice and implementation of the model, the tipe of learning, the creation and retrieval of data sets, and the obtained results. |
Secondary keywords: |
master theses;machine learning;deep learning;deep neural networks;recurrent neural networks; |
URN: |
URN:SI:UM: |
Type (COBISS): |
Master's thesis/paper |
Thesis comment: |
Univ. v Mariboru, Fak. za naravoslovje in matematiko, Oddelek za matematiko in računalništvo |
Pages: |
VIII, 71 f. |
ID: |
10949267 |