.png){kind=icon}
.png){kind=icon}
Return
Spike-train level supervised learning algorithm based on bidirectional modification for liquid state machines
.png){kind=icon}
PRE
AI
DOI：10.1007/s10489-022-04152-5.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Liquid state machine (LSM) of spiking neurons is a biologically plausible computational model imitating the structure and functions of the nervous system for information processing. Current supervised learning algorithms for spiking LSMs, such as the remote supervised method (ReSuMe), generally only adjust the synaptic weights in the output layer, while the synaptic weights of input and liquid layers are no longer changed during supervised learning. In this paper, a spike-train level supervised learning algorithm for spiking LSMs based on a bidirectional modification mechanism is proposed, which is called the bidirectional modification method (BiMoMe). Unlike ReSuMe, the proposed BiMoMe algorithm can adjust all synaptic weights in LSMs, which can enhance the dynamics of the network and is a biologically plausible supervised learning algorithm. The learning performance of BiMoMe is evaluated through several spike train learning tasks and an image classification problem. Experimental results show that BiMoMe has stronger spike train learning ability and image classification performance compared with other related algorithms, indicating that BiMoMe is effective in solving spatio-temporal pattern learning problems.
Keywords:
Supervised learning
Liquid state machines
Bidirectional modification
Synaptic plasticity
Journal
IF:
3.5
Papers:
7.5K
Citations:
1.7W