We have studied the self-organisation of the visual nervous system of mammals by neural network modelization [1].
We have modeled the development of ocularity domains in the visual cortex during the embryonic stage [2]. Our model is a self-organising neural network with two input layers (one for each retina) and one output layer representing the cortex. The connectivity between the layers is driven by Hebbian learning. Diffusion of activity between proximal neurons in the same layer and restriction terms in the weight growth are sufficient for explaining the development of ocularity domains. We have studied the geometry of the distribution of ocularity domains and its dependence on the parameters of the system.
Following the previous work, we modeled the development of neurons selective to the size and orientation of visual stimuli using a two layer neural network with: (i) diffusion of neural signal in the same layer, (ii) Hebbian and anti-Hebbian learning, and (iii) individual restrictions to the growth of each neuronal connection [3]. We analysed the resulting neuronal map.
The steady state of the model was used to simulate retinal plasticity in the adult visual cortex that is known to occur in response of a retinal lesion (scotoma). The scotoma was simulated as a perturbation of the input layer properties of the system [4]. The system autonomously evolves to a new steady state that displays properties in agreement with experimental results. The reorganization of the intra-cortical connectivity is shown as major responsible of the changes in receptive field positions.
We have extended the model presented in [3] to a binocular one in order to account for the simultaneous development of retinotopy, oriented receptive fields, and ocular dominance [5].
References
[1] Andrade, M.A. Doctoral thesis (in Spanish). March, 1994. Modelización del desarrollo ontogenético del sistema nervioso visual mediante redes neuronales. Universidad Complutense de Madrid. [PDF at UCM]
[2] Andrade, M.A. and F. Morán. 1996. Structural Study of the development of ocularity domains using a neural network model. Biological Cybernetics. 74, 243-254
[3] Andrade, M.A. and F. Morán. 1997. Receptive field map development by anti-Hebbian learning. Neural Networks. 10, 1037-1052
[4] Andrade, M.A., E.M. Muro and F. Morán. 2001. Simulation of plasticity in the adult visual cortex. Biological Cybernetics. 84, 445-451
[5] Muro, E.M., M.A. Andrade and F. Morán. 2002. A self-organizing model for the development of orientation selectivity and ocular dominance patterns in the visual nervous system. Neural Network World. 12, 319-332.