1: \begin{abstract}

2: We are motivated by problems that arise in a number of applications such as Online Marketing and explosives detection, where the observations are usually modeled using Poisson statistics. We model each observation as a Poisson random variable whose mean is a sparse linear superposition of known patterns. Unlike many conventional problems observations here are not identically distributed since they are associated with different sensing modalities. We analyze the performance of a Maximum Likelihood (ML) decoder, which for our Poisson setting involves a non-linear optimization but yet is computationally tractable. We derive fundamental sample complexity bounds for sparse recovery when the measurements are contaminated with Poisson noise. In contrast to the least-squares linear regression setting with Gaussian noise, we observe that in addition to sparsity, the scale of the parameters also fundamentally impacts sample complexity. We introduce a novel notion of Restricted Likelihood Perturbation (RLP), to jointly account for scale and sparsity. We derive sample complexity bounds for $\ell_1$ regularized ML estimators in terms of RLP and further specialize these results for deterministic and random sensing matrix designs.

3: %Convergence of $l_1$ regularized ML estimation is provided based on RLP as well as the so-called Restricted Eigenvalue condition. Our results apply to both deterministic and random sensing matrices and we present several results for both cases.  

4: %We show that when the sensing matrix satisfies the so-called Restricted Eigenvalue (RE) condition and its elements are bounded, the $\ell^1$ regularized ML decoder is consistent. Moreover, it converges exponentially fast in terms of number of observations. Our results apply to both deterministic and random sensing matrices and we present several results for both cases.  

5: \end{abstract}

6: