Abstract
Real data tensors are typically high dimensional; however, their intrinsic information is preserved in low-dimensional space, which motivates the use of tensor decompositions such as Tucker decomposition. Frequently, real data tensors smooth in addition to being low dimensional, which implies that adjacent elements are similar or continuously changing. These elements typically appear as spatial or temporal data. We propose smoothed Tucker decomposition (STD) to incorporate the smoothness property. STD leverages smoothness using the sum of a few basis functions; this reduces the number of parameters. An objective function is formulated as a convex problem, and an algorithm based on the alternating direction method of multipliers is derived to solve the problem. We theoretically show that, under the smoothness assumption, STD achieves a better error bound. The theoretical result and performances of STD are numerically verified.