{"id":"arxiv/2605.14860","url":"https://arxiv.org/abs/2605.14860","tags":["math.OC","cs.LG"],"score":0,"title":"A Non-Monotone Preconditioned Trust-Region Method for Neural Network Training","domain":"math.OC","source":"arxiv","authors":["Andrea Angino","Bindi Çapriqi","Shega Likaj","Ken Trotti","Rolf Krause"],"reviews":0,"subject":"math.OC","summary":"Training deep neural networks at scale can benefit from domain decomposition, where the network is split into subdomains trained in parallel and coupled by a global trust-region mechanism. Building on the Additively Preconditioned Trust-Region Strategy (APTS), we propose a non-monotone variant with a nonlinear additive Schwarz preconditioner that combines parallel subdomain corrections with global coarse-space directions. A windowed acceptance criterion allows controlled objective increases, avoiding needless rejection of effective coarse steps. The resulting non-monotone APTS (NAPTS)…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-05-14","reactions":{}}