{"id":"arxiv/2609.32441","url":"https://arxiv.org/abs/2609.32441","tags":["math.OC"],"score":0,"title":"Saddle-Point Problems with a Low-Dimensional Block Do Not Need Accurate Inner Solves","domain":"math.OC","source":"arxiv","authors":["Ivan Fomin","Alexander V. Gasnikov"],"reviews":0,"subject":"math.OC","summary":"Many learning problems couple a high-dimensional block of parameters with a handful of adversarial or dual variables: worst-group risk over a few groups, or learning under a few constraints. We consider $\\min_{x\\in X}\\max_{y\\in Y} f(x,y)$ with $X\\subseteq\\mathbb{R}^n$, a compact convex set and strongly convex with condition number $κ$, and we count the oracle calls for $x$ and for $y$ separately. The textbook approach runs a cutting-plane method in $y$ and solves every inner problem to accuracy $\\varepsilon$ with an accelerated method; it needs $O(m\\log(1/\\varepsilon))$ calls for $y$ but…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-09-26","reactions":{}}