Parallel simulation of spiking neural networks
Spiking neural networks transfer information through activation spikes that carry information through their weight and temporal delay. The behavior of a spiking neural network can be simulated through discrete event simulation, where neurons are framed as discrete logical processes. However, due to...
Saved in:
| 主要作者: | |
|---|---|
| 其他作者: | |
| 格式: | Final Year Project |
| 語言: | English |
| 出版: |
Nanyang Technological University
2021
|
| 主題: | |
| 在線閱讀: | https://hdl.handle.net/10356/153506 |
| 標簽: |
添加標簽
沒有標簽, 成為第一個標記此記錄!
|
| 總結: | Spiking neural networks transfer information through activation spikes that carry information through their weight and temporal delay. The behavior of a spiking neural network can be simulated through discrete event simulation, where neurons are framed as discrete logical processes. However, due to the large size of the network and the number of update events occurring, sequential discrete event simulation is unable to simulate a large population of spiking neurons effectively.
Application of parallel discrete event simulation techniques to simulate spiking neural networks allows training using biologically plausible learning rules on a large-scale platform that can be potentially decentralized. In this report, a prototype for GPU-based discrete-event simulation of spiking neurons is explored, demonstrating convergence and unsupervised learning using GPU-based algorithms. |
|---|