1: \begin{abstract} 

2:   This paper puts forth a novel bi-linear modeling framework for data recovery

3:   via manifold-learning and sparse-approximation arguments and considers its

4:   application to dynamic magnetic-resonance imaging (dMRI). Each temporal-domain

5:   MR image is viewed as a point that lies onto or close to a smooth manifold,

6:   and landmark points are identified to describe the point cloud concisely. To

7:   facilitate computations, a dimensionality reduction module generates

8:   low-dimensional/compressed renditions of the landmark points. Recovery of

9:   high-fidelity MRI data is realized by solving a non-convex minimization task

10:   for the linear decompression operator and affine combinations of

11:   landmark points which locally approximate the latent manifold geometry. An

12:   algorithm with guaranteed convergence to stationary solutions of the

13:   non-convex minimization task is also provided. The aforementioned framework

14:   exploits the underlying spatio-temporal patterns and geometry of the acquired

15:   data without any prior training on external data or information. Extensive

16:   numerical results on simulated as well as real cardiac-cine

17:   MRI data illustrate noteworthy improvements of the advocated machine-learning

18:   framework over state-of-the-art reconstruction techniques.

19: \end{abstract}

20: