Learning Permutation from Structure Without Supervision
For researchers working on unsupervised permutation learning (e.g., sorting, jigsaw reconstruction), this provides a practical improvement over fixed-temperature Gumbel-Sinkhorn, though the contribution is incremental.
The paper tackles the problem of learning permutations from structure without supervision, where existing methods using a single global temperature for Gumbel-Sinkhorn lead to instability. The proposed entropy-adaptive formulation improves training stability and final permutation quality, especially for larger problems with higher ambiguity.
Many learning problems require uncovering a hidden ordering that reveals structure in unordered data, such as monotonicity in sorting or spatial continuity in jigsaw reconstruction. In these settings, permutations can be learned as latent operators by optimizing objectives defined directly on the reordered output, often without access to ground-truth orderings. Differentiable relaxations such as Gumbel-Sinkhorn make this approach practical by approximating permutation matrices with doubly stochastic matrices. However, learning from structure without supervision induces a non-uniform uncertainty: some assignments become confident early, while others remain ambiguous. Existing methods control this process using a single global temperature, forcing all assignments to sharpen or diffuse simultaneously and leading to instability at scale. We introduce an entropy-adaptive formulation of Gumbel-Sinkhorn that locally modulates temperature based on assignment uncertainty. This allows confident assignments to discretize early while preserving exploration where uncertainty remains. Across sorting and jigsaw reconstruction tasks and in routing-style settings, adaptive entropy control improves training stability and final permutation quality relative to fixed-temperature baselines, particularly as problem size and assignment ambiguity increase.