A statistical perspective on algorithm unrolling models for inverse problems