GLISSOM is based on the RF-LISSOM computational model (Receptive-Field Laterally Interconnected Synergetically Self-Organizing Map; figure 1) [10, 11].RF-LISSOM focuses on the two-dimensional topographic organization of the cortex, modeling a cortical area as an N × N sheet of neurons and the retina as an R × R sheet of ganglion cells. Neurons receive afferent connections from broad patches of radius rA on the retina, and receive lateral excitatory and inhibitoryconnections from nearby and more distant neurons in the cortex (radii rE and rI, respectively). The connection weights are initially random or isotropic, and are subsequently organized through an unsupervised Hebbian learning process using visual input. Weak connections are eliminated periodically, resulting in patchy lateral connectivity similar to that observed in the visual cortex. Due to the large number of connections involved in a realistic map, this self-organization process is very computation and memory intensive.
Keywords for this software
References in zbMATH (referenced in 6 articles )
Showing results 1 to 6 of 6.
- Gauci, Jason; Stanley, Kenneth O.: Autonomous evolution of topographic regularities in artificial neural networks (2010)
- Toussaint, Marc: A sensorimotor map: modulating lateral interactions for anticipation and planning (2006)
- Choe, Yoonsuck; Miikkulainen, Risto: Contour integration and segmentation with self-organized lateral connections (2004)
- Lücke, Jörg: Hierarchical self-organization of minicolumnar receptive fields (2004) ioport
- Bednar, James A.; Miikkulainen, Risto: Learning innate face preferences (2003)
- Bednar, James A.; Kelkar, Amol; Miikkulainen, Risto: Modeling large cortical networks with growing self-organizing maps (2002)