{"id":"arxiv/2609.31380","url":"https://arxiv.org/abs/2609.31380","tags":["math.OC"],"score":0,"title":"Optimization with Region-Reduced ReLU Neural Networks","domain":"math.OC","source":"arxiv","authors":["Christoph Plate","Caroline Ganzer","Mirko Hahn","Alexander Klimek","Heyuan Liu","Sebastian Sager","Kai Sundmacher","Hanna Wilhelm"],"reviews":0,"subject":"math.OC","summary":"Optimization of mathematical models involving integer decisions and neural networks with ReLU activation (ReLU ANNs) is a challenging task. Nevertheless, such models are an enabling technology in many application domains. A prominent example is superstructure optimization in chemical engineering, where ReLU ANNs are frequently employed as surrogate models for complex nonlinear processes. We survey recent developments in this area. We argue that in addition to network size and training options of the ANNs, the ReLU activation geometry and the number of linear regions on the domain of interest…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-09-25","reactions":{}}