Sijbring, Daan (2020) Fine-Tuning Input Selection while Learning Associative Memories in a Spiking Neural Network. Master's Thesis / Essay, Human-Machine Communication.
|
Text
Master_Thesis-FINAL.pdf Download (7MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
toestemming.pdf Restricted to Registered users only Download (96kB) |
Abstract
This research aims at designing a spiking neural network capable of learning associative memories. An important effect in associative recognition tasks is the fan effect – increased error rates and reaction times and decreased neural activity for memory items which are associated with other memory items. Our goal is to investigate whether the fan effect occurs in the results of the network, and how influencing the network’s encoder selectivity through different Voja rule adaptations best approaches some expected manifestations of the fan effect. Learning the associative memory in a spiking neural network occurs through a combination of supervised Prescribed Error Sensitivity (PES) learning in connection weights and unsupervised Vector Oja (Voja) learning focusing on input selectivity. Measurements will involve the network’s accuracy of representation after training, and its predicted neural activity for certain inputs based on the trained input selectivity that is caused by the Voja rule (or adaptation thereof). Results will show that all examined Voja adaptations more closely exhibit the fan effect than the original Voja rule, and shows how further subtleties in the individual rules influence memory performance. The research ultimately presents ideas on how to further any input selectivity rules in order to better approach the fan effect, and reflects on what the newly studied Voja adaptations can contribute to associative memory learning in different applications.
| Item Type: | Thesis (Master's Thesis / Essay) |
|---|---|
| Supervisor name: | Borst, J.P. and Taatgen, N.A. |
| Degree programme: | Human-Machine Communication |
| Thesis type: | Master's Thesis / Essay |
| Language: | English |
| Date Deposited: | 10 Sep 2020 10:04 |
| Last Modified: | 10 Sep 2020 10:04 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/23387 |
Actions (login required)
 |
View Item |