An Efficient and Flexible Spike Train Model Via Empirical Bayes

Accurate statistical models of neural spike responses can characterize the information carried by populations. But limited samples counts during recording usually result in model overfitting. Besides, current assume to be Poisson-distributed, which ignores fact that many neurons demonstrate over-dispersed spiking behaviour. Although Negative Binomial Generalized Linear Model (NB-GLM) provides a powerful tool for modeling counts, maximum likelihood-based standard NB-GLM leads highly variable and inaccurate parameter estimates. Thus, we propose hierarchical parametric empirical Bayes method estimate among neuronal population. Our integrates both Models (GLMs) theory, aims (1) improve accuracy reliability estimation, compared Poisson-GLM; (2) effectively capture over-dispersion nature from simulated data experimental data; (3) provide insight into interactions behaviours We apply our approach study data. The estimation simulation indicates new framework accurately predict mean different recover connectivity weights based on retinal proposed outperforms Poisson-GLM terms predictive log-likelihood held-out 1