A GPU Parallel Algorithm for Non Parametric Tensor Learning

One rich source of large data sets is the high dimensionality of the data formats known as tensors. Compared to the vector use, learning with tensors is inherently more complex and requires high-performance computing. The aim of this paper is to investigate tensor-based algorithms for regression and classification, i.e. tensor learning, that are suitable to be implemented in parallel architecture to handle large data sets. To this end a tensor learning model based on a general theoretical framework for approximating a generic tensor function has been established. Then a parallel version of the model has been derived to benefit the GPU resources. Finally, extensive experiments on large data sets that use both CPU and GPU have been carried out to validate the proposed approach.