Collaborators: Rhys Bowden, UniMelb, Matthew Roughan (UoA)
Internet Tomography is the name given to a class of statistical inference problems where Internet measurements made over accessible `slices' of the network are used to infer quantities over other, inaccessible slices. Classic examples include measuring aggregrate traffic at routers inside the network to infer end-to-end traffic demand, or measuring end-to-end delays at measurement stations at the edge of the network to infer delays at nodes inside it.
This project researches a number of Internet Tomography problems. The main focii currently are:
Exploiting sparsity for loss tomography (in a nutshell, working out how to predict loss `hotspot' locations assuming there aren't many of them).
Investigating joint compressibility for topology inference (working out how the network is connected by observing the similarities in the structure of observed packet losses at different locations).
