Unfolding representations of trajectory coding in neuronal population spike activity

A fundamental goal of computational neuroscience is to understand the representation of large-scale neuronal population codes during various behavioral states. To date, a few probabilistic models have been developed for uncovering representations of neuronal population spike activity. In this paper, we propose an unsupervised learning framework for unfolding sequence representations in neuronal population codes. Specifically, we integrate temporal embedding into a hidden Markov model (HMM) and use it to unfold trajectory coding representations in neuronal population spike activity. Finally, we apply the proposed methods to two rat hippocampal recordings. Our preliminary results suggest that temporal embedding enhances the representation power of HMM-type models and improves the detection power for identifying significant hippocampal memory replay events during off-line state.