uncertainty-weighted aggregation

To address the high computational cost arising from joint propagation of aleatoric and epistemic uncertainties in Bayesian two-stage inference, this paper proposes an efficient uncertainty propagation framework. First, a representative subset is selected via Pareto-smoothed importance sampling to reduce sampling redundancy. Second, an importance-weighted moment-matching strategy is introduced for lightweight posterior approximation. Third, an iterative mixture-distribution expansion mechanism is developed to jointly model both uncertainty types within surrogate modeling and MICE-based multiple imputation. The method preserves posterior accuracy while significantly reducing the cost of multi-model fitting—achieving several-fold improvements in computational efficiency. It establishes a scalable paradigm for Bayesian inference under complex, heterogeneous uncertainty scenarios.

Uncertainty quantification in missing data imputation is often overlooked, and the relationship between calibration quality and imputation accuracy remains poorly understood. This paper presents the first systematic empirical evaluation of six state-of-the-art imputation methods—statistical (MICE, SoftImpute), distribution-alignment (OT-Impute), and deep generative (GAIN, MIWAE, TabCSDI)—across multiple real-world datasets, under MCAR, MAR, and MNAR missingness mechanisms, and across varying missing rates. We propose a multi-path evaluation framework integrating repeated sampling variability, conditional distribution modeling, and predictive confidence quantification to rigorously assess uncertainty calibration. Results reveal that high imputation accuracy does not imply well-calibrated uncertainty estimates; significant trade-offs exist among accuracy, calibration fidelity, and computational efficiency across method categories. We identify several robust, reproducible configurations, providing actionable, evidence-based guidance for model selection in downstream machine learning and data cleaning tasks.

This work proposes a model-agnostic framework that systematically incorporates uncertainty from missing values in photovoltaic (PV) power data into data-driven short-term forecasting. By leveraging stochastic multiple imputation combined with Rubin’s rules, the approach explicitly models and propagates missing-data uncertainty into the predictive distribution—a critical aspect often overlooked in existing methods. The framework effectively mitigates the overconfidence commonly observed in prediction intervals that fail to account for such uncertainty, thereby yielding substantially better-calibrated interval forecasts. Notably, this improvement in calibration is achieved without compromising the accuracy of point predictions, demonstrating a robust balance between reliability and precision in PV forecasting under incomplete data conditions.

In wind power probabilistic forecasting, sensor missing data are pervasive; however, the prevailing “impute-then-predict” paradigm introduces estimation bias and fails to propagate missingness uncertainty. To address this, we abandon the two-stage approach and propose an end-to-end framework based on joint generative modeling: input features and target variables are jointly modeled as a single distribution, and missing features are marginalized out via Bayesian integration—ensuring faithful uncertainty propagation. Our method employs expressive generative models—such as conditional variational autoencoders or normalizing flows—that natively handle incomplete inputs and implicitly learn the missingness mechanism during training. Experiments demonstrate that our approach achieves statistically significant improvements in Continuous Ranked Probability Score (CRPS) over diverse imputation-based baselines. Moreover, it incurs substantially lower inference overhead than existing uncertainty-aware forecasting methods, delivering superior accuracy, robustness to missingness patterns, and computational efficiency.

Missing values in multivariate time series—particularly electronic health records (EHRs)—often arise from sensor disconnections, yet existing imputation methods neglect model uncertainty, hindering reliable identification of low-confidence estimates. Method: We propose the first general-purpose imputation framework integrating uncertainty quantification: it employs probabilistic modeling techniques (e.g., Monte Carlo Dropout) to compute per-value confidence scores and introduces an adaptive threshold–driven selective imputation mechanism that fills only missing entries with sufficiently high confidence. Contribution/Results: Evaluated across multiple real-world EHR datasets, our approach significantly reduces imputation error and improves downstream clinical prediction performance—e.g., 24-hour mortality prediction—demonstrating both the efficacy and clinical utility of uncertainty-aware imputation.

This study addresses the challenge of uncertainty quantification in aggregated time series forecasting, particularly for annual totals and year-over-year growth rates. It proposes a simulation-augmented multi-step split conformal prediction method (SA-MSCP), which generates future trajectories via block bootstrap resampling from cross-validated residuals and constructs calibrated prediction intervals using empirical quantiles. By innovatively integrating a simulation-augmentation mechanism into the multi-step split conformal prediction framework, the method significantly improves empirical coverage for both aggregate totals and their growth rates, yielding more reliable uncertainty estimates without compromising predictive accuracy.

This work addresses the problem of aggregating multiple calibrated Bayesian expert forecasts to construct a new predictor that remains calibrated and is Blackwell-dominated by the target expert, rather than merely minimizing a specific loss function. Under the setting where only the experts’ prior distributions are observed—without access to the true state—the authors formally define the aggregation objective as simultaneously achieving calibration and Blackwell refinability. By modeling calibrated experts through reduced-form information structures, they characterize the set of feasible predictions using the row space of a linear system intersected with a non-negative cone, and analyze it via Blackwell dominance theory. Their main contributions include efficient solvability of both randomized aggregation problems, while showing that determining the existence of a deterministic aggregator is NP-hard and admits no multiplicative PTAS unless P = NP, thereby revealing a fundamental computational distinction between randomized and deterministic aggregation.

This work addresses the performance degradation in online forecasting of irregular multivariate time series caused by dynamically shifting data distributions. To tackle this challenge, the authors propose Under-Cali, a lightweight and model-agnostic online calibration framework that operates without updating the frozen source model. Under-Cali leverages uncertainty estimation as a core control signal and employs a dual-expert calibration mechanism coupled with an adaptive routing strategy to differentially process and jointly update samples with high versus low uncertainty. Evaluated across multiple benchmark datasets, the framework consistently achieves stable performance improvements while maintaining low computational overhead.