Leferink, Petra (2023) Transfer Learning in Spiking Neural Networks. Master's Thesis / Essay, Artificial Intelligence.
|
Text
mAI_2023_LeferinkP.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 (152kB) |
Abstract
Over the past decade, spiking neural networks coupled with event-based sensors have shown the potential for constructing energy-efficient and low-latency embedded systems. Existing embedded systems often employ fixed models from prior-to-deployment training, which can cause performance degradation from a shift of data distribution between domains due to noise, environmental conditions or user specificities. To mitigate this, transfer learning is employed, which involves initial training in a source domain to grasp common knowledge that exists across domains, followed by fine-tuning in a target domain. This study specifically explores the effectiveness of transfer learning with Spiking Neural Networks (SNNs) in speech recognition tasks. The focus lies on speech processed with artificial cochleas, where differences in information density across domains can arise from different temporal resolutions. To delve into the dynamics of transfer learning, we explored the impact of SNN, training-related, and data-processing parameters on the performance. The results show that transferring from a source domain with a greater information density to a less information-dense target domain outperforms the opposite transfer learning scenario. Furthermore, smaller differences between domains yield better final performance. The neural membrane potential time constant influences the performance of the model greatly, as we observed that a model with smaller neural membrane potential time constants
| Item Type: | Thesis (Master's Thesis / Essay) |
|---|---|
| Supervisor name: | Chicca, E. and Taatgen, N.A. and Fabre, M.P.Y. |
| Degree programme: | Artificial Intelligence |
| Thesis type: | Master's Thesis / Essay |
| Language: | English |
| Date Deposited: | 15 Jan 2024 08:50 |
| Last Modified: | 15 Jan 2024 08:50 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/31811 |
Actions (login required)
 |
View Item |