Hi folks --
I'm hoping that your collective wisdom can help me avoid a lot of false starts. Can anybody direct me to current best-practices regarding how to fit a model to spike-train data?
The gist of the problem is this: usually one is not trying to match precise spike times, but rather to fit higher-order phenomena such as spike rate adaptation or underlying slow-wave envelopes (like a T-current burst), in which the presence of spikes is important but their precise timing is not. Spikes are big phenomena, though, and most timeseries-processing optimization algorithms cannot help but weigh their precise timing heavily. Phase independent algorithms such as that included with Neurofitter fix this problem for uniform time offsets in data timeseries, but not for variability within the timeseries.
Sometimes one will be able to provide a set of neuronal responses to nominally identical stimuli so that the real-world variance in spike timing can be directly estimated. Sometimes it's just a less sophisticated desire to deemphasize spike timing with respect to other features in the timeseries. Any leads will help; I imagine a problem of this scope and relevance has been worked on by a lot of people.
Thank you all in advance for your help and thoughts.