Accelerating low-rank matrix completion on GPUs

Latent factor models formulate collaborative filtering as a matrix factorization problem. However, matrix factorization is a bi-linear problem with no global convergence guarantees. In recent years, research has shown that the same problem can be recast as a low-rank matrix completion problem. The resulting algorithms, however, are sequential in nature and computationally expensive. In this work we modify and parallelize a well known matrix completion algorithm so that it can be implemented on a GPU. The speed-up is significant and improves as the size of the dataset increases; there is no change in accuracy between the sequential and our proposed parallel implementation.