Loading...
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 51112 (05)
- University: Sharif University of Technology
- Department: Electrical Engineering
- Advisor(s): Shamsollahi, Mohammad Bagher
- Abstract:
- Transfer Learning is one of the most important fields in the Machine Learning area. Respect to the advances that we have seen in the Computer Science, especially in the Machine Learning area, we need a tool that can transfer learnings from different domains to each other. As data distribution varies, many statistical models require restructuring using new training data. In many applications, re-assembling training data and re-structuring models is inefficient and costly, so reducing the need for this practice seems appropriate. In these cases, knowledge transfer or learning transfer between domains may be desirable. For example, in the area of the B rain-Computer Interface, when it comes to classify the data for a given person, this classification can not be used on another person, meaning that the accuracy will be not as good as training a new classifier. The goal of transfer learning is to improve the designed classifier in the source domain to fit it into the target domain. Our goal in this project is to implement the methods of boosting based transfer learning to be used on the brain-computer interface datasets and improve some of currently proposed algorithms. In this study, we use two datasets, Decmeg 2014 datasets as well as BCI III 3A datasets. In the first chapter, the transfer learning was defined and various types of its models were introduced. After that, in the research background chapter, we reviewed the previous works on transfer learning strategies and the research works done in the field of Transfer Learning for brain-computer Interface. In the next chapter, we proposed a new algorithm on a multi- source basis by different codings, since the boosting based transfer learning algorithm is based on binarty classification. To do this, we provide different codings such as binary coding, One-hot coding, OOC and Random coding. In the results chapter, we first showed that, by increasing the length of the code, random coding leads to better classify multi-class topic than the One Against All method. Subsequently, we looked at Decmeg 2014 and BCI III 3a datasets and showed that, firstly, for the BCI III 3A datasets,since the winner has reached the accuracy of 84.6 percent, we have succeeded that accuary using proposed algorithm, Random coding and SVM as base classifier, increasing this amount to 90.6 percent. Also for the Decmeg 2014 datasets, since the winner's accuracy has reached to 77 percent, we reached to 83.4 percent using random coding and our proposed algorithm using SVM as base classifier
- Keywords:
- Transfer Learning ; Boosting Machine Learning ; Adaboost Algorithm ; Brain-Computer Interface (BCI)
Digital Object List
-
محتواي کتاب
- view
Bookmark