Paper compares predict-then-optimize and generate-then-optimize decisions
Original title: Contextual Decision Making: Predictive or Generative
Regret bounds show smoothness can offset generate-then-optimize's extra dimensions, while nonnegative-weight methods face a smoothness limit.
The paper compares predict-then-optimize, which learns a conditional expected objective, with generate-then-optimize, which learns conditional uncertainty and optimizes generated scenarios. Regret bounds indicate smoothness differences can offset the extra dimensional cost of the generative approach. Within nonnegative-weight predict-then-optimize methods, a lower bound limits gains from more smoothness, and signed weights can lose convexity preservation. Synthetic experiments and a case study illustrate tradeoffs; the authors do not name a universal winner.
- Generate-then-optimize keeps scenario-wise convexity.
- Flexible predict-then-optimize methods may use signed weights.
- Advice depends on data, problem structure, and model reuse.
- No approach is identified as always preferable.
From the original article
1 min read at arxiv.orgContextual decision making uses observed covariates to guide decisions under uncertainty. Two natural approaches are predict-then-optimize (PtO), which learns the conditional expected objective, and generate-then-optimize (GtO), which learns conditional uncertainty and optimizes using generated scenarios.
By L. Jeff Hong · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.