Toward a Biologically Plausible Model of LGN-V1 Pathways Based on Efficient Coding

Abstract

Increasing evidence supports the hypothesis that the visual system employs a sparse code to represent visual stimuli, where information is encoded in an efficient way by a small population of cells that respond to sensory input at a given time. This includes simple cells in primary visual cortex (V1), which are defined by their linear spatial integration of visual stimuli. Various models of sparse coding have been proposed to explain physiological phenomena observed in simple cells. However, these models have usually made the simplifying assumption that inputs to simple cells already incorporate linear spatial summation. This overlooks the fact that these inputs are known to have strong non-linearities such the separation of ON and OFF pathways, or separation of excitatory and inhibitory neurons. Consequently these models ignore a range of important experimental phenomena that are related to the emergence of linear spatial summation from non-linear inputs, such as segregation of ON and OFF sub-regions of simple cell receptive fields, the push-pull effect of excitation and inhibition, and phase-reversed cortico-thalamic feedback. Here, we demonstrate that a two-layer model of the visual pathway from the lateral geniculate nucleus to V1 that incorporates these biological constraints on the neural circuits and is based on sparse coding can account for the emergence of these experimental phenomena, diverse shapes of receptive fields and contrast invariance of orientation tuning of simple cells when the model is trained on natural images. The model suggests that sparse coding can be implemented by the V1 simple cells using neural circuits with a simple biologically plausible architecture.

Learning receptive field properties of complex cells in V1

“…One candidate mechanism to solve this problem is efficient coding, which can be 116 implemented in a biologically plausible fashion, through Hebbian plasticity, to explain 117 many experimental phenomena of simple cells [33]. However, we found that efficient 118 coding cannot effectively learn the RF properties of complex cells (see Discussion and S1 119 Appendix for details).…”

“…Parameters are defined in Table 1. A summary of the parameters of the model that will be used throughout this paper 134 is given in Table 1. and has been previously shown to account for many experimental phenomena [33]. This 137 two-layer model is a variant of sparse coding and incorporates many biological 138 constraints.…”

“…v S leak represents the change of 147 membrane potential caused by leakage currents. Details of the bottom two layers can be 148 found in [33]).…”

“…In addition, the absolute value of 220 each weight is limited to an upper bound, a 1,max , that represents the maximal synaptic 221 efficacy. The only difference between the previous learning rules of simple cells [33] and 222 this study is that weight normalization is replaced by the combination of The Bienenstock, Cooper, and Munro (BCM) rule [34,35] can learn underlying features 231 from the input through a competitive process between synapses of a neuron that arises 232 from the thresholding mechanism that is part of the BCM learning rule. However, BCM 233 plasticity is designed for single neurons and does not introduce any competition between 234 network neurons that receive the same visual input.…”

“…The learning rate, η 1 , is 3. 10 5 epochs are used 318 in the training process, similar to previous studies [33,52]. After the training process for 319 simple cells, the connections between LGN and simple cells, A u and A d , are fixed.…”