Optimal Transport Framework Unifies Cold-Start Active Learning Methods

A new transport selection framework unifies disparate cold-start active learning heuristics through an optimal transport lens. The method uses entropic regularization to control selection trade-offs and establishes a task-agnostic minimax bound for selecting unlabeled data subsets without prior knowledge.