Minimax-Optimal Aggregation for Density Ratio Estimation

ICLR 2026 Conference SubmissionAnonymous Authors
Density Ratio EstimationSelf-ConcordanceKernel MethodsDomain Adaptation
Abstract:

Density ratio estimation (DRE) is fundamental in machine learning and statistics, with applications in domain adaptation and two-sample testing. However, DRE methods are highly sensitive to hyperparameter selection, with suboptimal choices often resulting in poor convergence rates and empirical performance. To address this issue, we propose a novel model aggregation algorithm for DRE that trains multiple models with different hyperparameter settings and aggregates them. Our aggregation provably achieves minimax-optimal error convergence without requiring prior knowledge of the smoothness of the unknown density ratio. Our method surpasses cross-validation-based model selection and model averaging baselines for DRE on standard benchmarks for DRE and large-scale domain adaptation tasks, setting a new state of the art on image and text data.

Disclaimer
This report is AI-GENERATED using Large Language Models and WisPaper (A scholar search engine). It analyzes academic papers' tasks and contributions against retrieved prior work. While this system identifies POTENTIAL overlaps and novel directions, ITS COVERAGE IS NOT EXHAUSTIVE AND JUDGMENTS ARE APPROXIMATE. These results are intended to assist human reviewers and SHOULD NOT be relied upon as a definitive verdict on novelty.
NOTE that some papers exist in multiple, slightly different versions (e.g., with different titles or URLs). The system may retrieve several versions of the same underlying work. The current automated pipeline does not reliably align or distinguish these cases, so human reviewers will need to disambiguate them manually.
If you have any questions, please contact: mingzhang23@m.fudan.edu.cn

Overview

Taxonomy

Core-task Taxonomy Papers
33
3
Claimed Contributions
28
Contribution Candidate Papers Compared
3
Refutable Paper

Research Landscape Overview

Core task: density ratio estimation with model aggregation. The field encompasses a diverse set of methodological branches that address how to estimate ratios of probability densities and how to combine multiple models or estimators to improve performance. At the highest level, the taxonomy distinguishes core density ratio estimation methods—which develop direct techniques for computing ratios without separately estimating each density—from model aggregation and ensemble techniques that focus on combining predictions from multiple base learners. Additional branches cover density estimation with aggregation (where ensembles are applied to density estimation itself), domain adaptation and transfer learning (which leverage density ratios to handle distribution shift), Bayesian optimization and sequential decision making (where density ratios guide exploration), statistical inference and hypothesis testing (using ratios for two-sample tests or covariate shift correction), and specialized applications ranging from speech recognition to species distribution modeling. Representative works such as Ensemble Direct Density[9] and Density Ratio SuperLearner[11] illustrate how aggregation strategies can be tailored to density ratio problems, while methods like MBORE[4] show the utility of ratios in active learning settings. Within the model aggregation branch, a particularly active line of research investigates minimax-optimal aggregation strategies that provide theoretical guarantees on the combined estimator's performance. Minimax Density Ratio[0] sits squarely in this subfield, emphasizing rigorous risk bounds and optimal weighting schemes when aggregating density ratio estimators. This contrasts with more heuristic ensemble approaches seen in works like Ensemble Kernel Matching[22] or Bagged Copula GP[32], which prioritize empirical flexibility over worst-case optimality. A central trade-off across these aggregation methods is between computational tractability and the strength of theoretical guarantees: some techniques achieve near-oracle performance under mild assumptions, while others sacrifice formal optimality for broader applicability or ease of implementation. The original paper's focus on minimax optimality places it among a small cluster of theoretically driven aggregation studies, distinguishing it from the many empirical ensemble methods and from branches that apply density ratios to downstream tasks without emphasizing aggregation theory.

Claimed Contributions

Novel model aggregation algorithm for density ratio estimation

The authors introduce a new aggregation method that combines multiple density ratio estimators trained with different hyperparameters. The method optimizes aggregation weights by minimizing an upper bound on the Bregman divergence, yielding an analytic solution that is computationally efficient.

10 retrieved papers
Can Refute
Minimax-optimal convergence rates without prior smoothness knowledge

The authors prove that their aggregation approach achieves minimax-optimal error convergence rates for a broad class of DRE methods optimized in reproducing kernel Hilbert spaces, without needing to know the smoothness of the true density ratio in advance.

8 retrieved papers
Can Refute
Theory-grounded aggregation method addressing hyperparameter choice

The authors develop a principled aggregation framework that addresses the sensitivity of density ratio estimators to hyperparameter selection, providing both theoretical guarantees and practical improvements over cross-validation-based model selection.

10 retrieved papers

Core Task Comparisons

Comparisons with papers in the same taxonomy category

Within the taxonomy built over the current TopK core-task papers, the original paper is assigned to a leaf with no direct siblings and no cousin branches under the same grandparent topic. In this retrieved landscape, it appears structurally isolated, which is one partial signal of novelty, but still constrained by search coverage and taxonomy granularity.

Contribution Analysis

Detailed comparisons for each claimed contribution

Contribution

Novel model aggregation algorithm for density ratio estimation

The authors introduce a new aggregation method that combines multiple density ratio estimators trained with different hyperparameters. The method optimizes aggregation weights by minimizing an upper bound on the Bregman divergence, yielding an analytic solution that is computationally efficient.

Contribution

Minimax-optimal convergence rates without prior smoothness knowledge

The authors prove that their aggregation approach achieves minimax-optimal error convergence rates for a broad class of DRE methods optimized in reproducing kernel Hilbert spaces, without needing to know the smoothness of the true density ratio in advance.

Contribution

Theory-grounded aggregation method addressing hyperparameter choice

The authors develop a principled aggregation framework that addresses the sensitivity of density ratio estimators to hyperparameter selection, providing both theoretical guarantees and practical improvements over cross-validation-based model selection.