A Simulation Study on Neural Ensemble Sparse Coding

The aim of this paper is to investigate a new method to code neural ensemble activity via sparse theory, and the validity of sparse coding is shown by simulation study on coding a stimulus via neural ensemble. Experimental data used in this study is a simulation spatiotemporal spikes of 120 neurons in hippocampus CA3 computed from small-world neural network model with two types of stimuli: a pulse current and Gaussian white noise. Sparse coding is based on a probabilistic model which is able to find sparse linear decompositions of a data. The sparse coding algorithm is achieved by minimizing the reconstruction error with that of sparseness. In this research, the simulation data is preprocessed by a 100 ms wide window. The window is moved with step of 25 ms to establish a spatiotemporal distribution mapping. Then the data is decomposed by sparse coding. The results show that the outputs are sparse, which indicate that sparse theory can better describe the characteristic of neural ensemble. And the results also show that firing patterns of neural ensembles under the pulse current stimulus and Gaussian white noise stimulus are also different, which indicate that sparse coding is able to code different stimuli. So the validity of the neural ensemble sparse coding method has been proved.