| Form of studies |
Bachelor |
| Title of the study programm |
Computer Systems |
| Title in original language |
Normalizācijas funkciju salīdzinājums regresijas uzdevumos, izmantojot dziļo māšīnmācīšanos |
| Title in English |
Comparison of normalization functions in regression tasks using deep learning |
| Department |
Faculty Of Computer Science Information Tehnology And Energy |
| Scientific advisor |
Ēvalds Urtāns |
| Reviewer |
Marina Uhanova |
| Abstract |
Normalization is a widely used method in deep learning field. It allows to
increase performance of the learning process, performed by neural networks. The
bachelor thesis consists of empirical analysis of 3 existing normalization functions.
Experiments were performed based on combination of different values of
hyperparameters, together with 3 datasets of different size and complexity with respect
to features (Weather in Szeged, CalCOFI, Stellar classification). Batch normalization
function “LayerNorm” has shown best performance on average among all of the
combination of experiments, while function “BatchNormLast” was classified as the
most unstable and had lowest performance.
The bachelor thesis contains: 59 pages, 34 figures and 26 referenced
information sources. |
| Keywords |
PADZIĻINĀTA MĀCĪŠANĀS, PARTIJU NORMALIZĀCIJA, REGRESIJA |
| Keywords in English |
DEEP LEARNING, BATCH NORMALIZATION, REGRESSION |
| Language |
eng |
| Year |
2022 |
| Date and time of uploading |
29.05.2022 15:52:46 |
