{"id":"arxiv/2604.18540","url":"https://arxiv.org/abs/2604.18540","tags":["math.AP","cs.LG","math.FA","math.OC"],"score":0,"title":"Duality for the Adversarial Total Variation","domain":"math.AP","source":"arxiv","authors":["Leon Bungert","Lucas Schmitt"],"reviews":0,"subject":"math.AP","summary":"Adversarial training of binary classifiers can be reformulated as regularized risk minimization involving a nonlocal total variation. Building on this perspective, we establish a characterization of the subdifferential of this total variation using duality techniques. To achieve this, we derive a dual representation of the nonlocal total variation and a related integration of parts formula, involving a nonlocal gradient and divergence. We provide such duality statements both in the space of continuous functions vanishing at infinity on proper metric spaces and for the space of essentially…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-04-20","reactions":{}}