New Boltzmann margin achieves near-exponential kNN convergence

A new paper from researchers Luyuan Yang, Shayan Shafaei, and Chao Lan introduces a novel condition called the Boltzmann margin for analyzing classification convergence rates. Traditionally, classifiers are analyzed under either the Tsybakov margin (weak, polynomial rates) or the Massart margin (strong, exponential rates). The Boltzmann margin bridges this gap: it is weaker than Massart but generally stronger than Tsybakov, and can imply many properties of both under suitable conditions. The authors apply this margin to k-nearest neighbor (kNN) classifiers and prove, for the first time, near-exponential convergence rates for kNN classification. This is a significant theoretical advance, as kNN is a simple and widely used nonparametric method but has historically only achieved polynomial convergence guarantees.

The paper, accepted at the Conference on Uncertainty in Artificial Intelligence (UAI) 2026, also provides extensions of the main results and numerical evidence supporting the theoretical claims. The Boltzmann margin opens a new middle ground in margin theory, potentially allowing exponential convergence guarantees for other classifiers that previously only had polynomial bounds. For the machine learning community, this work refines our understanding of when and why certain algorithms can learn faster, offering both theoretical insight and practical guidance for designing more efficient learning systems.