1: \begin{abstract}
2: We study the problem of estimating a temporally varying coefficient
3: and varying structure (VCVS) graphical model underlying
4: nonstationary time series data, such as social states of interacting
5: individuals or microarray expression profiles of gene networks, as
6: opposed to {\it i.i.d.} data from an invariant model widely
7: considered in current literature of structural estimation. In
8: particular, we consider the scenario in which the model evolves in a
9: piece-wise constant fashion. We propose a procedure that minimizes
10: the so-called TESLA loss (i.e., temporally smoothed L1 regularized
11: regression), which allows jointly estimating the partition
12: boundaries of the VCVS model and the coefficient of the sparse
13: precision matrix on each block of the partition. A highly scalable
14: proximal gradient method is proposed to solve the resultant convex
15: optimization problem; and the conditions for sparsistent estimation
16: and the convergence rate of both the partition boundaries and the
17: network structure are established for the first time for such
18: estimators.
19: \end{abstract}
20: