Combining Lateral and Elastic Interactions: Topology-Preserving Elastic Nets
作者:Valery Tereshko, Nigel M. Allinson
摘要
We have developed a topology-preserving elastic net which combines both lateral and synaptic interactions to obtain topologically ordered representations (receptive fields) of an external stimulus. Existing neural models that preserve the topology by utilizing lateral interactions, such as the Kohonen-type maps, and by utilizing synaptic interactions, such as elastic-type nets, appear as limiting cases of this model. Utilizing both lateral and synaptic interactions can be beneficial for preservation of the topology.
论文关键词:cortical map, deterministic annealing, elastic net, free energy, Kohonen map, soft competition, topology-preserving map, unsupervised learning
论文评审过程:
论文官网地址:https://doi.org/10.1023/A:1015721115089