Central European University, Austria
Network reconstruction from indirect observations
The observed functional behavior of a wide variety large-scale systems is often the result of a network of pairwise interactions. However, in many cases these interactions are hidden from us, either because they are impossible or very costly to be measured directly, or, in the best case, are measured with some degree of uncertainty. In such situations, we are required to infer the network of interactions from indirect information.
In this talk, I present a scalable Bayesian method to perform network reconstruction from indirect data, including noisy measurements and observed network dynamics. This kind of approach allows us to convey in a principled manner the uncertainty present in the measurement, and combined with versatile modeling assumptions can yield good results even when data are scarce. In particular, I describe how the reconstruction approach can be combined with community detection, allowing us to tap into multiple sources of evidence available for the task. I show how this combined approach provides a twofold improvement, by increasing not only the reconstruction accuracy, but also the identifiability of communities in networks. The latter improvement is possible even in situations where at first we might imagine that reconstruction is impossible, for example when we the network has been measured only once and we lack any kind of error assessment.