Abstract
We present an approach to distinguish between causal connections and common input connections among nodes in a network. By modeling how the activity of a node depends on its own recent history, we demonstrate how this history dependence predicts different patterns of activity depending on the nature of the network connectivity. In particular, a causal connection between a pair of observed nodes can be distinguished from common input connections that originate from nodes whose activity remains unobserved. This work builds on previous results where this same distinction was made based on modeling how the activity of a node depends on measured external variables such as stimuli. The results have a potentially broad range of application as the analysis can be based on a fairly generic class of models.
Original language | English (US) |
---|---|
Pages (from-to) | 354-391 |
Number of pages | 38 |
Journal | SIAM Journal on Applied Mathematics |
Volume | 68 |
Issue number | 2 |
DOIs | |
State | Published - 2007 |
Keywords
- Autocorrelation
- Causality
- Correlations
- Maximum likelihood
- Neural networks
- Point process