Recent progress in recording techniques now allows researchers to record from hundreds and even thousands of neurons simultaneously. New, scalable methods need to be developed to handle such large data sets. Ideally these methods should not only analyse the multi-dimensional data, but also provide results which could be interpreted by the brain.

One key problem in neuronal data analysis is to identify neuronal assemblies i.e. groups of neurons displaying coordinated neuronal activity. Currently available methods either search for assemblies whose neurons participate in repeated spike sequences or search for assemblies whose neurons display similar firing rate modulations. In this thesis, I present two approaches to the search for neuronal assemblies.

I investigate whether a spiking neural network equipped with biologically plausible synaptic learning rules can provide a biologically interpretable way of finding repeating spike patterns in neuronal data. Due to the similarities between spiking neural networks to how brains function, this might be very close to how the brain itself detects such repeated activity.

Furthermore, I present neural topic modelling – a new data analysis method for large neuronal data sets. Based on a machine learning method from text mining, neural topic modelling is scalable and produces interpretable results. By including multiple features of neuronal spike trains and even other data types such as local field potentials into the analysis, I can expand the definition of neuronal assemblies to any type of neuronal activity features which are co-modulated. The application of neural topic modelling to neuronal recordings reveals interactions between features of neuronal activity which have previously not been identified.

Since both approaches are biologically plausible, the results from both methods can be used to generate hypotheses about how the brain processes information and may reveal hitherto unknown information processing pathways.