Characterize the benefits of MEQ iteration across problem types

Determine when and for what types of problems iterative refinement with the Mutual EQuilibrium (MEQ) architecture provides benefits.

Background

The paper demonstrates iterative refinement primarily through multimodal classification and visual-grounding experiments, but it does not establish which task properties make repeated mutual feedback useful. The authors therefore identify the conditions under which MEQ's iteration helps, and the classes of problems that benefit from it, as an unresolved future direction.