A manuscript has been submitted for this project and a pre-print is coming soon!
Figure caption: Our method clusters behavioral (1) and non-behavioral (0) in 2-dimensional reduction through the TSNE method. This approach outperforms prior methods (other boxes) that fail to classify odor clusters under a variety of different parameters.
We have developed a spatial-temporal attention based generative method to model the interarrival spike interval (ISI) of neuron spikes. Our generative method has demonstrated remarkable ability to model the distribution of ISI. Moreover, we have shown that spatio-temporal attention, specifically the spatial weights assigned to each neuron, may encode the stimulus type. In our studies, it was found that the spatial weights indicate a specific set of neurons important for the modeling of the ISI of behavioral stimuli. Indeed, using the TNSE method for 2D reduction, we clustered spatial attention weights into two clusters and observed that these correspond to behavioral and non-behavioral stimuli. We are extending this technique (using spatio-temporal attention) to classify neuron types. Note that current recording techniques and methods are unable to identify specific neuron types.