In light of the increased interest in tensors, here is an implementation I hadn't mentioned before:
A kernel-based framework to tensorial data analysis by Marco Signoretto, Lieven De Lathauwer, Johan A.K. Suykens
Tensor-based techniques for learning allow one to exploit the structure of carefully chosen representations of data. This is a desirable feature in particular when the number of training patterns is small which is often the case in areas such as biosignal processing and chemometrics. However, the class of tensor-based models is somewhat restricted and might suffer from limited discriminative power. On a different track, kernel methods lead to flexible nonlinear models that have been proven successful in many different contexts. Nonetheless, a naïve application of kernel methods does not exploit structural properties possessed by the given tensorial representations. The goal of this work is to go beyond this limitation by introducing non-parametric tensor-based models. The proposed framework aims at improving the discriminative power of supervised tensor-based models while still exploiting the structural information embodied in the data. We begin by introducing a feature space formed by multilinear functionals. The latter can be considered as the infinite dimensional analogue of tensors. Successively we show how to implicitly map input patterns in such a feature space by means of kernels that exploit the algebraic structure of data tensors. The proposed tensorial kernel links to the MLSVD and features an interesting invariance property; the approach leads to convex optimization and fits into the same primal–dual framework underlying SVM-like algorithms.
An implementation of the examples of the paper can be found here.
This image was taken by Navcam: Right B (NAV_RIGHT_B) onboard NASA's Mars rover Curiosity on Sol 976 (2015-05-06 01:37:54 UTC).
Image Credit: NASA/JPL-Caltech
This image was taken by Navcam: Right B (NAV_RIGHT_B) onboard NASA's Mars rover Curiosity on Sol 976 (2015-05-06 01:37:54 UTC).
Image Credit: NASA/JPL-Caltech
No comments:
Post a Comment