Astral Space: Convex Analysis at Infinity

2022/05/06 by Dudík, Miroslav, Schapire, Robert E., Telgarsky, Matus · 1 citation
#FOS: Computer and information sciences #FOS: Mathematics #Machine Learning (cs.LG) #Optimization and Control (math.OC)

paper · doi:10.48550/arxiv.2205.03260

Abstract

Not all convex functions on ℝⁿ have finite minimizers; some can only be minimized by a sequence as it heads to infinity. In this work, we aim to develop a theory for understanding such minimizers at infinity. We study astral space, a compact extension of ℝⁿ to which such points at infinity have been added. Astral space is constructed to be as small as possible while still ensuring that all linear functions can be continuously extended to the new space. Although astral space includes all of ℝⁿ, it is not a vector space, nor even a metric space. However, it is sufficiently well-structured to allow useful and meaningful extensions of concepts of convexity, conjugacy, and subdifferentials. We develop these concepts and analyze various properties of convex functions on astral space, including the detailed structure of their minimizers, exact characterizations of continuity, and convergence of descent algorithms.

Cited by

• How Transformers Learn In-Context Recall Tasks? Optimality, Training Dynamics and Generalization

Related

• Convex analysis on Hadamard spaces and scaling problems
• Extreme points and geometric aspects of compact convex sets in asymmetric normed spaces
• Basic convex analysis in metric spaces with bounded curvature
• The finest locally convex topology of an extended locally convex space
• Convexity of sets and quadratic functions on the hyperbolic space