Neuronal populations in vivo are characterized by the recurring coactivation of specific groups of neurons, referred to as neuronal assemblies. These assemblies are thought to constitute fundamental units for brain computations. Identifying neuronal assemblies through statistical analysis of simultaneous recordings of large neuronal populations is crucial. Here, we describe a computationally fast algorithm based on dimensionality reduction techniques, developed to detect neuronal assemblies and analyze their dynamics. Importantly, it allows for overlap between the detected assemblies, where neurons are able to transiently participate in multiple assemblies. We show that this method has been successfully applied for the analysis of calcium imaging experiments involving thousands of simultaneously recorded neurons, and its accuracy and scalability is supported by benchmarking studies on simulated data.