Tensor Completion

Tensor completion aims to recover missing entries in a multi-dimensional data array (tensor) by exploiting its underlying low-rank structure. Current research emphasizes developing efficient algorithms, often based on tensor decompositions like CP or Tucker, and exploring various sampling strategies beyond simple Bernoulli sampling to improve computational efficiency and accuracy. These advancements are crucial for handling large-scale datasets with missing values in diverse applications, including recommender systems, image/video inpainting, and network latency estimation, where accurate data completion is essential for reliable analysis and prediction. Furthermore, research is actively addressing challenges such as handling noisy data, incorporating prior knowledge (e.g., graph information), and ensuring fairness and consistency in the completed tensor.