Our goal is to develop scalable analytical methods to capture hidden dynamics of network activity and to arrive at biologically meaningful interpretations of massive datasets. This work is done in collaboration with Prof. Schwing's group.

​

       Since complex behavior is a result of coordinated firing in a large neuronal network, we expect the recorded activity to exhibit a rich hidden structure. In a search for complex high dimensional  representations, we develop structured machine learning approaches capable of learning the dynamics of the recorded neural network as a whole. Subsequent dimensionality reduction and correlation with behavioral and sensory datasets allows us to extract functional neural circuits and systems involved in complex behavior. Instead of analyzing the neural data based on a pre-conceived biological model, our approach enables efficient reduction to low-dimensional dataset in a model-unbiased manner.