Abstract:We establish strong duality relations for functional two-step compositional risk-constrained learning problems with multiple nonconvex loss functions and/or learning constraints, regardless of nonconvexity and under a minimal set of technical assumptions. Our results in particular imply zero duality gaps within the class of problems under study, both extending and improving on the state of the art in (risk-neutral) constrained learning. More specifically, we consider risk objectives/constraints which involve real-valued convex and positively homogeneous risk measures admitting dual representations with bounded risk envelopes, generalizing expectations and including popular examples, such as the conditional value-at-risk (CVaR), the mean-absolute deviation (MAD), and more generally all real-valued coherent risk measures on integrable losses as special cases. Our results are based on recent advances in risk-constrained nonconvex programming in infinite dimensions, which rely on a remarkable new application of J. J. Uhl's convexity theorem, which is an extension of A. A. Lyapunov's convexity theorem for general, infinite dimensional Banach spaces. By specializing to the risk-neutral setting, we demonstrate, for the first time, that constrained classification and regression can be treated under a unifying lens, while dispensing certain restrictive assumptions enforced in the current literature, yielding a new state-of-the-art strong duality framework for nonconvex constrained learning.

Duet expectile preferences

Disappointment Concordance and Duet Expectiles

Dual Representation of Expectile-Based Expected Shortfall and Its Properties

Probabilistic risk aversion for generalized rank-dependent functions

A Rank-Dependent Theory for Decision under Risk and Ambiguity

Numerical Representation of Convex Preferences Over Anscombe-Aumann Acts

Expectations-Based Loss Aversion May Help Explain Seemingly Dominated Choices in Strategy-Proof Mechanisms

Strong Duality in Risk-Constrained Nonconvex Functional Programming

Expectations-based Loss Aversion in Contests

A theory for combinations of risk measures

The Neyman–Pearson lemma for convex expectations

Robust Utility Maximization with Extremely Ambiguity-Loving and Ambiguity-Aversion Preferences

Maxmin expected utility with non-unique prior

Loss Aversion and State-Dependent Linear Utility Functions for Monetary Returns

Strong Duality Relations in Nonconvex Risk-Constrained Learning

Decision-making Under Ordinal Preferences and Comparative Uncertainty

Mixture independence foundations for expected utility

Dual theory of choice with multivariate risks

Relative Bound and Asymptotic Comparison of Expectile with Respect to Expected Shortfall

An elementary proof of the dual representation of Expected Shortfall