Bervoets, Antoon (2025) Controlling Recurrent Neural Networks with Improved Feature Conceptors. Master's Thesis / Essay, Artificial Intelligence.
|
Text
mAI2025BervoetsAGB.pdf Download (782kB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
Toestemming.pdf Restricted to Registered users only Download (126kB) |
Abstract
Conceptors are used to control various neural network architectures. Matrix Conceptors can be used to control recurrent neural networks and feed forward neural networks. Although Conceptor Control yields good performance, each Matrix Conceptor’s size is equal to the number of neurons squared. Diagonal Conceptor matrices were introduced to address the high storage cost. However, they lack performance and flexibility of use. Random Feature Conceptors (RFCs) are an expansion of these diagonal conceptors. Using Principal Component Analysis (PCA) this research further improves the RFC architecture. Here we show that with this newly proposed PCA based method, stability and prediction accuracy improve compared to RFCs. These results are broadly applicable in most scenarios where matrix Conceptors and RFCs are used.
| Item Type: | Thesis (Master's Thesis / Essay) |
|---|---|
| Supervisor name: | Jaeger, H. and Pourcel, G. A. |
| Degree programme: | Artificial Intelligence |
| Thesis type: | Master's Thesis / Essay |
| Language: | English |
| Date Deposited: | 07 Mar 2025 12:34 |
| Last Modified: | 07 Mar 2025 12:34 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/34845 |
Actions (login required)
 |
View Item |