{"id":"arxiv/2610.28025","url":"https://arxiv.org/abs/2610.28025","tags":["math.ST","stat.ME","stat.ML"],"score":0,"title":"Optimal aggregation in multiclass classification","domain":"math.ST","source":"arxiv","authors":["Felix Abramovich","Or Davidovitch-Servadio"],"reviews":0,"subject":"math.ST","summary":"This paper studies the problem of aggregation in multiclass classification, generalizing existing results for binary classification. Given a finite dictionary of score functions F={f_1,...,f_M} and a training sample, the goal is to construct a data-driven aggregate whose risk is as close as possible to that of the best aggregate (oracle) in a prescribed subset of span(F) with respect to a given loss. In particular, we investigate model selection and convex aggregation under the 0/1 and multinomial logistic losses. For the 0/1 loss, we show that the optimal aggregation rates remarkably do not…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-10-20","reactions":{}}