- Rating
- 3
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 3
Read review
Paper summary
This paper proposes Generative Near-Optimal Policy Learning (GenNOP), a framework for human-centered decision-making in settings where algorithmic objectives are measurable but important aspects of human judgment are difficult to formalize. The central premise is that optimizing a measurable outcome alone may lead to decisions that are suboptimal with respect to an overall utility , where represents an implicit human-centered objective. The synthetic results demonstrate substantial improvements in near-optimal action coverage over several baselines.
Critical strengths
- A meaningful formulation of human–algorithm complementarity that avoids explicitly modeling implicit human preferences.
The most compelling aspect of the paper is its formulation of human-centered decision-making as a problem in which measurable performance and implicit human preferences are handled by different components.
- A concrete generative-learning formulation supported by synthetic results.
The paper goes beyond a conceptual proposal by developing a concrete learning procedure combining a conditional extreme-value filter, propensity-based reweighting, and generative modeling.
The derivation of the reweighted training distribution is particularly useful because it clarifies how observational data can be transformed into a distribution that emphasizes potentially near-optimal actions while reducing dependence on historical action assignments.
The synthetic results provide encouraging evidence. In Table 2, GenNOP substantially improves both precision and recall relative to the evaluated baselines. The ablations further demonstrate that the near-optimality filter is important, while the end-to-end simulations suggest that preserving diverse alternatives can improve decision quality when implicit preferences matter.
These results support the feasibility of the proposed approach and suggest that learning a distribution of acceptable actions, rather than a single optimal action, can be beneficial in appropriately structured decision problems.
Critical weaknesses
- A substantial gap remains between the theoretical results, the learned distribution, and the claimed human-centered decision objective.
The paper's central motivation is to improve overall utility through human selection among actions that are near-optimal in measurable performance . However, the theoretical analysis does not sufficiently establish this connection.
In particular, Proposition 3.2 establishes a local ordering relationship between and for candidates whose -values lie sufficiently close to a particular reference point. In Appendix E, this reference point is the -coordinate of the interior maximizer of overall utility . In contrast, Definition 3.3 and the proposed policy in Definition 4.1 define near-optimality relative to the maximum attainable measurable outcome .
These reference points are generally distinct under the paper's own assumptions. Indeed, Proposition 3.1 explicitly considers the case in which maximizing differs from maximizing .
Consequently, the local ordering result does not directly justify selecting the action with the largest from the -optimal set used by GenNOP. Moreover, the tolerance in Proposition 3.2 is sample-dependent, whereas the algorithm uses a practitioner-specified tolerance. The practical implication of the proposition is therefore limited.
As a consequence, the theoretical and algorithmic connections underlying the paper's central claim remain insufficiently established.
- The experimental evidence does not adequately establish the practical effectiveness of the proposed human-centered decision-making framework.
The synthetic experiments are promising, but several aspects limit their ability to substantiate the central claim.
First, the end-to-end evaluations simulate human judgment using explicitly defined functions for and . This is reasonable for controlled experimentation, but the core motivation of the paper concerns preferences that cannot readily be formalized. There is no evaluation involving actual human decision-makers, making it difficult to assess whether the proposed alternatives are useful, interpretable, or meaningfully improve human decisions.
Second, the preference-axis experiment in Figure 6 uses an oracle tolerance
which requires knowledge of the true overall-utility maximizer. The authors acknowledge that this is not a practical calibration procedure. Nevertheless, the dependence of performance on is substantial, as illustrated by Figure 5 and Appendix K.4. Since selecting this parameter is an essential part of the proposed framework, stronger evidence under realistically chosen tolerances is necessary.
Third, the high-dimensional evaluation relies on Fashion-MNIST with class-dependent outcomes. Although the generated actions are images, the optimization structure is determined primarily by a small number of discrete classes. Furthermore, recall is assessed through class-level coverage. This limits the evidence that GenNOP can effectively handle complex high-dimensional action spaces with genuinely high-dimensional outcome landscapes.
Finally, the real-data experiments provide relatively weak validation. The MIMIC-IV results are evaluated primarily through two-dimensional distributional visualizations rather than quantitative measures of near-optimality, decision quality, or human benefit.
Importantly, Appendix K.5 states that the clinical experiments use a conditional Gaussian-process generator with GEV-derived observation weights, rather than the conditional diffusion architecture emphasized in the proposed implementation. This distinction does not invalidate the general framework, but it limits the relevance of these experiments as evidence for the specific approach being proposed.
Additionally, the observational clinical data involve substantial causal-identification challenges. The selected covariates are limited, the analysis includes only patients who survived to discharge, and hospital length of stay is influenced by many factors beyond the recorded medication decisions. The authors appropriately avoid claiming clinical optimality, but this also means the real-data experiments offer little direct evidence for the central decision-making objective.
The current evaluation therefore establishes that GenNOP can generate plausible and diverse alternatives under selected conditions, but not convincingly that these alternatives improve human-centered decisions in realistic settings.
This weakness is consequential because the principal advantage claimed for the method is improved human–algorithm decision-making, whereas the strongest evidence concerns a surrogate distribution-learning task.
Questions for authors
- Proposition 3.2 establishes local consistency between the ordering induced by and the ordering induced by around the measurable-outcome coordinate of the overall-utility optimum. However, the -optimal action set used by GenNOP is defined around the maximum of .
How does Proposition 3.2 justify selecting the -maximizing action from the set defined in Definition 4.1? Can the authors establish a corresponding result for that set, with a nontrivial tolerance that does not depend on the realized candidate sample?
A rigorous clarification or a corrected guarantee would meaningfully strengthen the theoretical contribution.
- Evidence concerning practical tolerance selection and comparison with simpler alternatives.
Figure 6 relies on an oracle choice of , even though this information would be unavailable in the intended applications.
How does GenNOP perform when is selected solely from observable data and a predefined acceptable loss in , without reference to the ground-truth overall utility ?
Additionally, how does the method compare with a simpler baseline that first estimates expected outcomes and then applies threshold-based filtering and diversity-aware candidate selection?
Thank you for your review! Please see our responses below:
Sec. 3andApp. E, refined our Pareto-frontier and coverage assumptions (Assumps. E.1, E.4), and provedLemma E.6andProp. 3.4directly on the fixed tolerance window [y⋆(x)−ϵ,y⋆(x)]. InProp. 3.4(Sec. 3, p. 4), we decomposed finite-sample expected utility regret into three interpretable components: an under-exploration bias when ϵ is too small to reach yV⋆(x), a linear over-exploration outcome loss when ϵ exceeds the optimal trade-off, and an O(ϵ/m) finite-sample selection gap across the m generated candidates. Using this decomposition, we proved that for an interval of fixed ϵ>0 and sufficiently large m, human selection over ϵ-optimal candidates strictly dominates deterministic Y-maximization in expected overall utility V, theoretically explaining the U-shaped ϵ trade-off and O(m−1/2) convergence observed inFig. 5. We also addedProp. 4.2(Sec. 4.1, p. 6;App. H.2) bounding the total-variation distance between our importance-weighted surrogate target qϵ and the uniform ϵ-optimal policy πϵ⋆, and showing that stabilized propensity weighting preservesProp. 3.4's finite-sample guarantees with the near-frontier density constant scaled by κp(x)−1.Threshold+FPSComparison: We added observable-only ϵ calibration rules based on empirical outcome dispersion (standard-deviation, interquartile-range, and inter-quantile multiples of observed Y) inSec. 3(p. 5) and evaluated them over five seeds inSec. 5(Table 3) andApp. K.4(Tables 8–10,Fig. 13) alongsideThreshold+FPS,Hard Threshold+Uniform, and full hyperparameter grids forDDOM,AWR,GP-UCB, andPerturbat matched candidate budgets m. On the 1D multi-peak benchmark (Sec. 5,Table 3,Table 10), where unweighted outcome regression smooths over narrow secondary modes,GenNOP Pureat observable ϵ=q95−q75 achieves the lowest mean regret across all evaluated methods and significantly outperformsThreshold+FPS,DDOM,AWR, andIPTW Onlywhile matching oracleHard Threshold+Uniform. On the 2D Ackley benchmark (Sec. 5,Table 3,Tables 8–9,Fig. 13), observable rules forGenNOP Purematch Pure Oracle ϵ⋆ without statistically significant regret differences, andGenNOP+RFat observable 1.00SD(Y) significantly outperformsGP-UCB,Perturb, andPoint Policywhile achieving lower mean regret than grid-bestDDOMandAWR; full comparisons across all threshold baselines and hyperparameter grids appear inTable 3andTables 8–10. Moreover, as demonstrated on 784-D Fashion-MNIST (Sec. 5,Table 2), amortized conditional diffusion scales natively to high-dimensional continuous action spaces where grid-based thresholding and farthest-point sampling become impractical.Sec. 5andApp. K.3that Fashion-MNIST uses class-archetype outcome mappings solely as an evaluation oracle to quantify multi-modal mode coverage under confounding, while our conditional diffusion model conditions only on demographic covariates (without access to class labels) and generates directly in 784-D pixel space, achieving substantially lower FID and higher target recall than all baselines (Sec. 5,Table 2). Furthermore, we added a five-seed patient-grouped quantitative evaluation of the Section 4 conditional diffusion and VAE pipelines on thecardioandsepsisMIMIC-IV cohorts inSec. 5(Table 4) andApp. K.7(Tables 13–14,Fig. 16) using train-only Stage-1 estimators and dual gradient-boosted and random-forest outcome evaluators. Across both clinical cohorts,GenNOP (Diff + GPS + GEV)achieves high held-out ϵ-optimal precision and significantly improves precision and predicted LOS overBehavior CloningandGenNOP w/o GEV(as well as short-stay W1 oncardioand relative toAWRonsepsis), while preserving substantial intra-patient candidate diversity (Sec. 5,Table 4,Table 14).