Recently there has been a growing interest in diffusion and gossip based protocols that can compute simple aggregates, such as the average and
max/min, of data acquired and stored in a sensor network. In this talk, we show how distributed average consensus can be used to compute more
complex aggregations, in particular the global least-squares estimattion of a set of unknown parameters. In our setup, each sensor in the network takes linear measurements of the unknown parameters, corrupted by independent noises. Each sensor node can take a finite or infinite number
of measurements asynchronously. We propose a space-time diffusion scheme that allows every node to asymptotically compute the global least-square estimation. Moreover, at every iteration of the algorithm, each node obtains a local, unbiased intermediate estimate with improving quality.
We show that this algorithm is robust to unreliable communication links, and works in a network with dynamically changing topology.