Id: arxiv/2608.05680 Url: https://arxiv.org/abs/2608.05680 Tags: math.OC Score: 0 Title: Curvature Residual Geometry in Bregman Regression Domain: math.OC Source: arxiv Authors: Vu Khac Ky Reviews: 0 Subject: math.OC Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-08-06 Page: https://delphi.science/r/arxiv/2608.05680 We consider linear regression models fitted by minimizing Bregman losses of the form \[ \frac{1}{n}\sum_{i=1}^n \left[ φ(y_i)-φ(x_i^\topθ) -φ'(x_i^\topθ) \bigl(y_i-x_i^\topθ\bigr) \right], \] where \(y_i\) is the observed response and \(x_i^\topθ\) is the linear prediction. Even when the generating potential \(φ\) is strongly convex, the resulting regression objective may be nonconvex in \(θ\). This creates a gap between the convexity of the generating potential and the optimization geometry of the fitted model. The Hessian can be written as a weighted Gram matrix whose weights depend on the…