Robustness of Tensor Product

A rank 3 tensor product network has 3 dimensions of size p, q, and r, say. Its processing units include pqr Networks Using Distributed binding units, and input/output units grouped into 3 Representations vectors of length p, q and r. Typically one of the I/O vectors represents a predicate, and the other two represent William H. Wilson a pair of arguments. The information represented in the Computer Science and Engineering tensor (i.e. in the binding unit structure) is thus relational University of New South Wales information, such as larger–than(mare, foal). With rank 2 Sydney NSW 2052 Australia networks, a common use is to represent a single relation [email protected] (say larger–than), and the two axes of the tensor would thus represent the two items that, as a pair, belong to that Graeme S. Halford relation, e.g. (mare, foal). The relation may be Psychology specialized, as in [6] to a functional relation, e.g. binding University of Queensland between members of a set of variables and a set of values. Queensland 4072 Australia [email protected]