Title: Generalization of generative model for neuronal ensemble inference method

Abstract: Various brain functions that are necessary to maintain life activities materialize through the interaction of countless neurons. Therefore, it is important to analyze the structure of functional neuronal network. To elucidate the mechanism of brain function, many studies are being actively conducted on the structure of functional neuronal ensemble and hub, including all areas of neuroscience. In addition, recent study suggests that the existence of functional neuronal ensembles and hubs contributes to the efficiency of information processing. For these reasons, there is a demand for methods to infer functional neuronal ensembles from neuronal activity data, and methods based on Bayesian inference have been proposed. However, there is a problem in modeling the activity in Bayesian inference. The features of each neuron's activity have non-stationarity depending on physiological experimental conditions. As a result, the assumption of stationarity in Bayesian inference model impedes inference, which leads to destabilization of inference results and degradation of inference accuracy. In this study, we extend the expressivity of the model in the previous study and improve it to a soft clustering method, which can be applied to activity data with non-stationarity. In addition, for the effectiveness of the method, we apply the developed method to synthetic data generated by the leaky-integrate-and-fire model, and discuss the result.