Exploit low-rank factorizations of the attribution tensor

Develop low-rank Tucker or canonical-polyadic factorizations of the attribution tensor as models of bounded theory of mind, using factorization rank as a cognitive-capacity parameter.

Background

The paper represents interactive belief using a third-order attribution tensor whose indices identify the observing agent, the modeled agent, and the proposition under consideration. For most of the analysis, the proposition mode is treated as a stack of independent matrix slices, while genuinely multilinear effects are handled through matrix and Kronecker-product representations.

The authors explicitly identify a fuller mathematical exploitation of the tensor structure as an unresolved direction. In particular, low-rank Tucker or canonical-polyadic decompositions could provide a principled representation of bounded theory of mind, with rank serving as a parameter for cognitive capacity. This problem would extend the paper’s linear operator framework toward dimensionality-constrained models of higher-order social cognition.