Vectorial Dimension Reduction for Tensors Based on Bayesian Inference
This addresses the challenge of preserving high-order structures in tensors for applications like data analysis, though it appears incremental as it builds on CP decomposition with a Bayesian approach.
The paper tackles the problem of dimensionality reduction for high-order tensors by introducing a probabilistic model that directly reduces tensors to vectors using a linear combination of basis tensors, outperforming existing PCA-based methods and CP decomposition in classification and clustering accuracy on public databases.
Dimensionality reduction for high-order tensors is a challenging problem. In conventional approaches, higher order tensors are `vectorized` via Tucker decomposition to obtain lower order tensors. This will destroy the inherent high-order structures or resulting in undesired tensors, respectively. This paper introduces a probabilistic vectorial dimensionality reduction model for tensorial data. The model represents a tensor by employing a linear combination of same order basis tensors, thus it offers a mechanism to directly reduce a tensor to a vector. Under this expression, the projection base of the model is based on the tensor CandeComp/PARAFAC (CP) decomposition and the number of free parameters in the model only grows linearly with the number of modes rather than exponentially. A Bayesian inference has been established via the variational EM approach. A criterion to set the parameters (factor number of CP decomposition and the number of extracted features) is empirically given. The model outperforms several existing PCA-based methods and CP decomposition on several publicly available databases in terms of classification and clustering accuracy.