Ising-like model for neural representation of natural images

Efficient coding hypothesis was proposed as a model of sensory coding in the brain. One noteworthy assumption in efficient coding hypothesis is that the neuronal activities are independent. However, many neuroscience experiments evidences show that neuron interactions and correlations are ubiquitous in the retina and cortical vortex, independent hypothesis may be inappropriate in efficient coding. In this paper, Ising model is employed to describe pairwise correlation of the neural activities . A novel biological spike neuron network model based on efficient coding is proposed. When adapted to the statistics of natural images, our model can reproduce receptive fields resemble Sparse coding or Independent Components Analysis (ICA). The results validate the proposed modeling approach in this paper.