06月07日(Fri) 09:00〜11:00 H会場(-市民プラザ3F AVスタジオ)
| 演題番号 | 4H1-6 |
|---|---|
| 題目 | Self-sustaining and self-silencing activity in spiking neural networks |
| 著者 | Hubert Julien(東京大学) 池上 高志(東京大学) |
| 時間 | 06月07日(Fri) 10:20〜10:40 |
| 概要 | Spiking neural networks are a class of neural networks designed to model closely the brain activity. Like in real neurons, spiking neurons transfer information through discrete events called spikes where a fixed amount of electricity is sent to efferent neurons. A spike is emitted once the current received from other neurons reaches a threshold. This differs from rate based neural networks where the information transfer between neurons is continuous. In those networks, like in the brain, memory and learning were always considered as being synaptic in nature, i.e. learning was achieved through the modification of synapses by a process known as Hebbian learning. Recently, this traditional role of Hebbian learning has been challenged. In rate based neural networks, it has been shown that non-synaptic memory is achievable through the network dynamics. It has never been shown how to achieve the same result in a spike based neural network. In this work, we take one step in that direction by presenting a spiking neural network capable of, first, sustaining its activity without external stimulation and, second, going silent after a fixed predetermined amount of time. This can represent the retention and forgetting of a single unit of memory within the network, and offers an alternative to synaptic based memory models. |
| 論文 | PDFファイル |