Risk-sensitive constraints in contextual optimization

Develop contextual-optimization methods that place uncertainty in the constraint and impose a risk-sensitive criterion on the constraint tail, rather than restricting uncertainty to an objective evaluated under risk neutrality.

Background

The paper contrasts its formulation with contextual-optimization literature organized around uncertain objectives under risk neutrality. Its setting instead places uncertainty in a resource constraint and controls the upper tail of cumulative cost. The authors report that both this placement of uncertainty and the use of a risk-averse constraint criterion are directions identified as open in the cited survey.

References

Both placements are among the directions the survey identifies as open beyond its core of uncertain objectives under risk neutrality.

— Chance-constrained selection of sequential intervention strategies from counterfactual estimates  (2608.13209 - Kim et al., 13 Aug 2026) in Section 2.2, The predict-then-optimize line of work

Net active power provides a simple measure of the balance between load demand and PV generation. Other variables, such as recent voltages, forecasts, or control policy-action differences, could also be used to group the scenarios; identifying the most informative context variable is left for future work.

— Safety Screening for Voltage Control in Active Distribution Grids via Distributionally Robust Conformal Screening  (2608.30889 - Bouchkati et al., 31 Aug 2026) in Section 3, subsection “Context-Aware Calibration” (Section IV, subsection “Context-Aware Calibration” in the experimental discussion)

Several directions remain open for future work. First, the convergence analysis relies on conservative global Lipschitz constants and sampling assumptions. Developing weaker convergence conditions and adaptive local Lipschitz estimates could reduce conservatism. Approaches that avoid requiring a known Lipschitz constant altogether provide another promising direction . Second, the current certificates assume deterministic black-box evaluations of $q_i$. When $q_i$ is estimated from finite Monte Carlo samples or neural surrogates, statistical uncertainty and surrogate approximation error needs to be incorporated into the box bounds.

— Risk-Aware Optimal Control with Rulebooks  (2609.05199 - Wongpiromsarn, 4 Sep 2026) in Section Conclusions and Future Work