|
Abstract:
|
While temporal patterns of activity provide valuable insight into relationships among variables, time course observations are often insufficient for drawing causal conclusions. This is to a large extent because of two complicating factors, one very specific to time course observations, and the other a more general factor. First, causal effects among variables, if any, may occur at much finer time scale than the available measurements. In such cases, using the observed time series data could identify effects that are starkly different than the true causal effects. Second, inferring causal conclusions in the presence of unmeasured confounders is not feasible, and time course observations are not exempt from this rule. In this talk, we will discuss how the continuous nature of point process data offers a unique opportunity to overcome the first obstacle. We also discuss how data from common multi-experimental settings in neuroscience offer an opportunity to overcome the second obstacle. In addition to theoretical justifications, the validity of the proposed framework for causal inference is investigated using empirical data.
|
