Dear all, On behalf of Prof. Diego Garlaschelli, I am pleased to announce that next week *Professor Michel Mandjes *will visit IMT Lucca and give a seminar entitled *Inference Problems in Stochastic Networks *(see abstract below). The seminar will take place on: *March 18th, 2026 – 3:00 PM*San Francesco Complex – Sagrestia Join online at imt.lu/sagrestia. All the best, Anna ------------------------------------------------ *Abstract*: The majority of the random graph literature focuses on inherently static models, in which features of the graph are considered only at a single point in time. Yet there are strong practical motivations to study stochastically evolving graphs, which capture the inherent dynamics of real-world networks. Moreover, one can even consider stochastic processes evolving on these dynamic graphs, further enriching the modeling framework. The main focus of my talk is on estimating, in the framework discussed above, model parameters from partial information. For example, in a first basic variant, we demonstrate how the underlying parameters of a dynamic random graph can be inferred from snapshots of the subgraph counts. A second model is an age-structured branching process, the most elementary instance being a setting with just juveniles and adults. Remarkably, the model parameters can be estimated just observing the total population. Next, I consider a static network of nodes represented as infinite-server queues. Our goal is to estimate key parameters — such as arrival rates, service-time distributions, and the routing matrix — using observations of the network’s population vector at Poisson-sampled time points. We propose a method-of-moments estimator and establish its consistency. Numerical experiments show that this approach provides accurate estimates even in high-dimensional settings. We present two variants: one assuming a known parametric form for the service-time distributions, and a "model-free version" that does not rely on such assumptions. Finally, we study a population process evolving on a dynamic random graph. Using time series data on the number of individuals at each node, we successfully estimate parameters governing the graph dynamics. We prove that the estimator is asymptotically normal.