Data Completion

Matrix and tensor completion is the problem of filling in the missing entries of a partially observed array — the mathematics behind recommender systems, image inpainting, and sensor data recovery. When the underlying data is (approximately) low-rank, a small fraction of observed entries can be enough to reconstruct the whole thing.

This repository contains research code for low-rank matrix and tensor completion, connected to my PhD dissertation on geometric approaches to the completion problem. The image above shows a demonstration: a portrait reconstructed from a sparse set of observed pixels using low-rank completion.

See the repository for the algorithms and experiments.