Uncertainty Propagation via Optimal Transport Ambiguity Sets

Abstract

Uncertainty propagation has established itself as a fundamental area of research in all fields of science and engineering. Among its central topics stands the problem of modeling and propagating distributional uncertainty, i.e., the uncertainty about probability distributions. In this paper, we employ tools from Optimal Transport to capture distributional uncertainty via Optimal Transport ambiguity sets, which we show to be very natural and expressive, and to enjoy powerful topological, geometrical, computational, and statistical features and guarantees. We show that these ambiguity sets propagate nicely and intuitively through nonlinear, possibly corrupted by noise, transformations, and that in many cases the result of the propagation is again an Optimal Transport ambiguity set. Moreover, whenever this is not the case, we show that the result of the propagation can be tightly upper bounded by another Optimal Transport ambiguity set. Importantly, our methodology allows us to capture exactly how complex systems shape distributional uncertainty. To conclude, we exemplify our findings in three fundamental applications in forward propagation, inverse problems, and distributionally robust optimization.

Modeling of Political Systems using Wasserstein Gradient Flows

“…Our model captures the trend in the data, but it disregards impact factors such as the historical context, election rounds, and political campaigns. These aspects, together with further theoretic analysis (e.g., regularization), connections with dynamic game theory [31] and uncertainty propagation [32], and case studies (e.g., asymmetric initial ideological distributions), are possible avenues for future research.…”

Stochastic Wasserstein Gradient Flows using Streaming Data with an Application in Predictive Maintenance

“…Furthermore, the iterates not only convergence to a Wasserstein ball but also provide practically relevant information if the generated solution (µ k ) k∈N is subsequently used for prediction or estimation purposes. In particular, one can deploy recent results in uncertainty propagation (Aolaritei et al, 2022) to study how Wasserstein balls propagate through prediction processes or leverage the framework of distributionally robust optimization to evaluate the worst-case risk over Wasserstein balls (Mohajerin Esfahani and Kuhn, 2018; Blanchet and Murthy, 2019;Gao and Kleywegt, 2022).…”

Uncertain uncertainty in data-driven stochastic optimization: towards structured ambiguity sets

“…Its proof is given in Appendix B and it is based on appropriate modifications of the technical approach developed in [10]. 2 From Assumptions 3 and 4(i), it follows that for any function φ ∈ m U (Ξ; R∪{+∞}) the integral Ξ φ(ζ)dQ(ζ) is well defined. This ensures that the integral of the function φ λ (ζ) := sup ξ∈Ξ {h(ξ) − ⟨λ, c(ζ, ξ)⟩} in the attainable dual pair in Theorem 6.4 is also well defined, since φ λ is universally measurable (cf.…”

“…of propagating optimal transport ambiguity sets is considered in [13,14,2], which take into account multiple data assimilation nonidealities. Further applications of DRO include economic dispatch in power systems [42], congestion avoidance in traffic control [39], and motion planning in dynamic environments [33].…”