Tucker Decomposition Group: Expressive Vitality and Comparability
Authors: Ye Liu, Junjun Pan, Michael Ng
Abstract: Deep neural networks have achieved a superb success in fixing many machine finding out and laptop computer imaginative and prescient points. The precept contribution of this paper is to develop a deep group primarily based totally on Tucker tensor decomposition, and analyze its expressive power. It is confirmed that the expressiveness of Tucker group is further extremely efficient than that of shallow group. Often, it is required to utilize an exponential number of nodes in a shallow group to have the ability to signify a Tucker group. Experimental outcomes are moreover given to test the effectivity of the proposed Tucker group with hierarchical tensor group and shallow group, and present the usefulness of Tucker group in image classification draw back
