We will discuss two problems of network traffic inference with hidden data. The first is one of estimating average link delay times from end-to-end delay times. A model that introduces spatial dependence across links is used to improve upon traditional independence models.

The second problem is estimating link available bandwidth from end-to-end available bandwidth measurements. A statistical model is used that also includes dependence across links.

The methods are shown to work on simulated network data in some situations. Simulations show that the phenomenon of spatial dependence is quite significant in both contexts.