Millea, A. (2014) Explorations in Echo State Networks. Master's Thesis / Essay, Artificial Intelligence.
|
Text
Thesis_Adrian_Millea_FINAL.pdf - Published Version Download (3MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
akkoord_MilleaA.pdf - Other Restricted to Registered users only Download (102kB) |
Abstract
Echo State Networks are powerful recurrent neural networks that can predict time-series very well. However, they are often unstable, making the process of finding an ESN for a specific dataset quite hard. We will explore this process, by employing different versions of the activation function, different weight matrices and different topologies. We will show the close connection between the ESN and Compressed Sensing, a recent field in signal processing. Moreover, we will try to tackle some of the main problems in the ESN construction process: minimize the variability between different initializations of the weight matrix, automate the process of finding an ESN without the need for extensive manual trial-and-error sequences and finally eliminate noise from the activation function to increase precision and lower computational costs associated with it. A high level of performance is achieved on many time-series prediction tasks. We also employ the ESN to trade on the FOReign EXchange market using a basic trading strategy, and we achieve significantly more profit compared to previous research.
| Item Type: | Thesis (Master's Thesis / Essay) |
|---|---|
| Degree programme: | Artificial Intelligence |
| Thesis type: | Master's Thesis / Essay |
| Language: | English |
| Date Deposited: | 15 Feb 2018 07:57 |
| Last Modified: | 15 Feb 2018 07:57 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/11878 |
Actions (login required)
 |
View Item |