Parametrization of the spectrum in stochastic analytic continuation

We present progress in the stochastic sampling approach to numerical analytic continuation of imaginary-time dependent correlation functions computed with quantum Monte Carlo simulations. The spectral functions are parametrized as a sum of \delta-functions in the continuum with fixed amplitudes and adjustable locations. This form can be efficiently sampled and is a good starting point for introducing various constraints and additional features. We introduce a simple criterion for choosing the optimal temperature that leads to sampling of only statistically relevant spectral features (avoiding over-fitting). Further more, we demonstrate how various prominent spectral features, e.g., edges and isolated \delta-functions, can be incorporated with the sampling approach and lead to significantly better fidelity.