Pals, Remco (2024) Towards an Iterative Approach for Constructing Diagonal Conceptors for Autonomous Pattern Generation. Bachelor's Thesis, Artificial Intelligence.
|
Text
bAI2024Pals.pdf Download (2MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
toestemming.pdf Restricted to Registered users only Download (429kB) |
Abstract
Conceptors are a neuro-computational mechanism that allow for controlling the reservoir dynamics of recurrent neural networks for a variety of tasks. Diagonal conceptors are a lighter version of these conceptors that trade some of the accuracy and robustness for an increase in ease of computation and reduction in storage required. Steps were made towards an iterative approach to compute them, such that on-the-fly learning would be possible. It was found that the established training procedure is not suited for iteration due to unstable conceptor weights and consequently erratic reservoir dynamics. Additionally, diagonal conceptors were applied to a feedback network to produce temporal patterns autonomously. It was argued that this alternative setup is potentially more suitable for an iterative approach, and its limitations were discussed.
| Item Type: | Thesis (Bachelor's Thesis) |
|---|---|
| Supervisor name: | Jaeger, H. |
| Degree programme: | Artificial Intelligence |
| Thesis type: | Bachelor's Thesis |
| Language: | English |
| Date Deposited: | 24 Jul 2024 14:40 |
| Last Modified: | 24 Jul 2024 14:40 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/33677 |
Actions (login required)
 |
View Item |