Reviews: An inner-loop free solution to inverse problems using deep neural networks

An interesting paper that solves linear inverse problems using a combination of two networks: one that learns a proximal operator to the signal class of interest, and the other that serves as a proxy for a large scale matrix inversion. The proximal operator is reusable whenever the signal domain is unchanged. One would need to retrain only the matrix inversion network when the underlying problem is changed. This is a significant advantage towards reusability of the training procedure. Strengths Novelty A very interesting problem Weaknesses - An important reference is missing - Other less important references are missing - Bare-bones evaluation The paper provides an approach to solve linear inverse problems by reducing training requirements.