Institutional Repository of Vilnius University
Besikeičiančio uždavinio sprendimas naudojant daugiasluoksnį perceptroną /
Abstract
dc:descriptionThis paper analyses the use of multilayer perceptron in a changing task environment and how its learning parameters affect its effectiveness. Firstly, a conceptual analysis is done. Then, a changing task is defined and generated. A 2-2-1 multilayer perceptron is adapted to solve the defined task. Experiments are done by changing the learning rate eta, momentum and the number of training iterations. Analysis is done on how the multilayer perceptron adapts to a gradually changing task as opposed to a rapidly changing task. It is found, that higher learning rate and momentum values help train the multilayer perceptron better when there is less training data, but when there is plentiful training data lower learning rate and momentum values help achieve more accurate results. Moreover, it is found, that the multilayer perceptron retains the memory of previous training, therefore the results after training the multilayer perceptron with new data start to worsen. Even though the results worsen each time the multilayer perceptron is retrained, the results are still better by gradually changing the task rather than rapidly.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2016
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Kaniušonis, Vytautas,
- Contributors dc:contributor
-
- Valaitis, Vytautas
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD24177353&prefLang=en_US