Papers

Overparameterized models often generalize well even when they interpolate noisy training data. This is known as benign overfitting. For linear classification, Hashimoto et al. (2025) analyzed the phenomenon under a broad class of mixture distributions, but only for homogeneous classifiers without a bias term. We extend their framework to classifiers with an intercept. Benign overfitting still occurs, but the intercept perturbs the normalized Gram matrix of the noise and creates extra constraints on the covariance. These constraints are strongest with label noise. Their effect depends on the covariance: under isotropic noise they are dominated by the homogeneous conditions, while in anisotropic or noisy regimes they can raise the dimensionality needed for benign generalization. Thus the bias term changes the theory in some covariance regimes and leaves the asymptotic thresholds unchanged in others.

Sampling from a target distribution induced by training data is central to Bayesian learning, with Stochastic Gradient Langevin Dynamics (SGLD) serving as a key tool for scalable posterior sampling and decentralized variants enabling learning when data are distributed across a network of agents. This paper introduces DIGing-SGLD, a decentralized SGLD algorithm designed for scalable Bayesian learning in multi-agent systems operating over time-varying networks. Existing decentralized SGLD methods are restricted to static network topologies, and many exhibit steady-state sampling bias caused by network effects, even when full batches are used. DIGing-SGLD overcomes these limitations by integrating Langevin-based sampling with the gradient-tracking mechanism of the DIGing algorithm, originally developed for decentralized optimization over time-varying networks, thereby enabling efficient and bias-free sampling without a central coordinator. To our knowledge, we provide the first finite-time non-asymptotic Wasserstein convergence guarantees for decentralized SGLD-based sampling over time-varying networks, with explicit constants. Under standard strong convexity and smoothness assumptions, DIGing-SGLD achieves geometric convergence to an $O(\sqrt{\eta})$ neighborhood of the target distribution, where $\eta$ is the stepsize, with dependence on the target accuracy matching the best-known rates for centralized and static-network SGLD algorithms using constant stepsize. Numerical experiments on Bayesian linear and logistic regression validate the theoretical results and demonstrate the strong empirical performance of DIGing-SGLD under dynamically evolving network conditions.

Bayesian spatial modeling provides a flexible framework for whole-brain fMRI analysis by explicitly incorporating spatial dependencies, overcoming the limitations of traditional massive univariate approaches that lead to information waste. In this work, we introduce SIMBA, a Scalable Image Modeling using a Bayesian Approach, for group-level fMRI analysis, which places Gaussian process (GP) priors on spatially varying functions to capture smooth and interpretable spatial association patterns across the brain volume. To address the significant computational challenges of GP inference in high-dimensional neuroimaging data, we employ a low-rank kernel approximation that enables projection into a reduced-dimensional subspace. This allows for efficient posterior computation without sacrificing spatial resolution, and we have developed efficient algorithms for this implemented in Python that achieve fully Bayesian inference either within minutes using the Gibbs sampler or within seconds using mean-field variational inference (VI). Through extensive simulation studies, we first show that SIMBA outperforms competing methods in estimation accuracy, activation detection sensitivity, and uncertainty quantification, especially in low signal-to-noise settings. We further demonstrate the scalability and interpretability of SIMBA in large-scale task-based fMRI applications, analyzing both volumetric and cortical surface data from the NARPS and ABCD studies.

A finite-horizon variant of the quickest change detection (QCD) problem that is of relevance to learning in non-stationary environments is studied. The metric characterizing false alarms is the probability of a false alarm occurring before the horizon ends. The metric that characterizes the delay is latency, which is the smallest value such that the probability that detection delay exceeds this value is upper bounded to a predetermined latency level. The objective is to minimize the latency (at a given latency level), while maintaining a low false alarm probability. Under the pre-specified latency and false alarm levels, a universal lower bound on the latency, which any change detection procedure needs to satisfy, is derived. Change detectors are then developed, which are order-optimal in terms of the horizon. The case where the pre- and post-change distributions are known is considered first, and then the results are generalized to the non-parametric case when they are unknown except that they are sub-Gaussian with different means. Simulations are provided to validate the theoretical results.

Bayesian optimization (BO) has been widely used to optimize expensive and gradient-free objective functions across various domains. However, existing BO methods have not addressed the objective where both inputs and outputs are functions, which increasingly arise in complex systems as advanced sensing technologies. To fill this gap, we propose a novel function-on-function Bayesian optimization (FFBO) framework. Specifically, we first introduce a function-on-function Gaussian process (FFGP) model with a separable operator-valued kernel to capture the correlations between function-valued inputs and outputs. Compared to existing Gaussian process models, FFGP is modeled directly in the function space. Based on FFGP, we define a scalar upper confidence bound (UCB) acquisition function using a weighted operator-based scalarization strategy. Then, a scalable functional gradient ascent algorithm (FGA) is developed to efficiently identify the optimal function-valued input. We further analyze the theoretical properties of the proposed method. Extensive experiments on synthetic and real-world data demonstrate the superior performance of FFBO over existing approaches.

We introduce Conformal Online Learning of Koopman embeddings (COLoKe), a novel framework for adaptively updating Koopman-invariant representations of nonlinear dynamical systems from streaming data. Our modeling approach combines deep feature learning with multistep prediction consistency in the lifted space, where the dynamics evolve linearly. To prevent overfitting, COLoKe employs a conformal-style mechanism that shifts the focus from evaluating the conformity of new states to assessing the consistency of the current Koopman model. Updates are triggered only when the current model's prediction error exceeds a dynamically calibrated threshold, allowing selective refinement of the Koopman operator and embedding. Empirical results on benchmark dynamical systems demonstrate the effectiveness of COLoKe in maintaining long-term predictive accuracy while significantly reducing unnecessary updates and avoiding overfitting.

Forecasting many sparse series requires two choices: how much of each series' own past to retain, and how much to borrow from other series. We show that when one exponential recency operator is applied to item- and group-level statistics alike in a hierarchical empirical-Bayes hurdle model, the two become one decision: forgetting preserves the leverage of a shared prior while raising the noise floor against which fine cross-series structure must be resolved. We formalize this two-sided effect as a fitting-window diagnostic, a credibility bound and a resolution condition for candidate pools, and derive from it an ex ante screen for regimes where refinement cannot pay. On five public intermittent-demand panels comprising over 19,000 series, where item-level evidence is sparsest, the diagnostic is right in both directions. Where truncation lowers credibility the shared prior pays: switching it off costs $5\%$ and $12\%$ at the shortest histories on the two panels it helps most, against about $1\%$ at full length, and the gain sits in the occurrence block. On four panels the diagnostic reports at most $0.6\%$ of room for refinement and realized gains are no larger; on the one panel with room the learned mixture does not collect it, which locates the open problem in the partition objective. The resulting forecaster, EBB, stays within about $3\%$ of a per-series Gaussian-process specialist at fixed origin and within $1\%$ under walk-forward evaluation where both run, at two to three orders of magnitude lower cost.

Human migration exhibits complex spatiotemporal dependence driven by environmental and socioeconomic forces. Modeling such patterns at scale requires methods that accommodate many random effects while remaining feasible when raw data or large design matrices cannot be freely shared across distributed nodes. We develop a communication-efficient inference framework for Varying Coefficient Mixed Models (VCMMs) with flexible mean structures and large correlated random-effect components. Using a Bayesian hierarchical representation of penalized splines, we derive sufficient statistics that preserve each node's likelihood contribution and recover the estimator from the full data under unrestricted communication. Under communication constraints, these statistics support a one-step communication-efficient estimator with first-order efficiency. An SVD-enhanced implementation stabilizes large or ill-conditioned random-effect covariance operators. Theory establishes likelihood preservation, convergence, asymptotic efficiency, and finite-sample concentration. Simulations and U.S. migration-flow data demonstrate accuracy, scalability, and recovery of dynamic spatial patterns.

This paper introduces a concept for change-point detection based on normalized entropy as a fundamental metric, aiming to overcome the dependence of traditional entropy methods on assumptions about data distribution and absolute scales. Normalized entropy maps entropy values to the [0,1] interval through standardization, accurately capturing relative changes in data complexity. By utilizing a sliding window to compute normalized entropy, this approach transforms the challenge of detecting change points in complex time series, arising from variations in scale, distribution, and diversity, into the task of identifying significant features within the normalized entropy sequence, thereby avoiding interference from parametric assumptions and effectively highlighting distributional shifts. Experimental results show that normalized entropy exhibits significant numerical fluctuation characteristics and patterns near change points across various distributions and parameter combinations. The average deviation between fluctuation moments and actual change points is only 2.4% of the sliding window size, demonstrating strong adaptability. This paper provides theoretical support for change-point detection in complex data environments and lays a methodological foundation for precise and automated detection based on normalized entropy as a fundamental metric.

Cross-validation (CV) is routinely used across the sciences to select models and tune parameters, and the resulting choices are often interpreted as substantive scientific conclusions (e.g., which variables, mechanisms, or risk factors are “supported by the data”). A key part of the CV procedure – the hold-out size, or equivalently the fold count $K$ – is typically set by convention (e.g., 80/20, $K=5$) rather than by a principled criterion. Central to the issue is the tradeoff between training and testing: increasing the training sample size improves model accuracy, while sacrificing certainty around the accuracy itself. We formalize the tradeoff by targeting predictive performance and explicitly penalizing evaluation uncertainty, which cannot be identified from the data without additional assumptions. We derive finite-sample expressions of this evaluation uncertainty under symmetric errors and general upper bounds under broader error conditions, yielding a transparent utility-based rule for selecting the hold-out size as a function of an irreducible-noise parameter. Empirical analyses with linear regression and random forests across multiple domains, and a high-dimensional genomics application, show that (i) the choice of $K$ is dependent on the data and model. (ii) the optimal $K$ varies based on the assumption on the irreducible error, and (iii) the implied inferential conclusions can change materially as the irreducible error, and thus $K$, varies. The resulting framework replaces a one-size-fits-all convention with a context-specific, assumption-explicit choice of $K$, enabling more reliable model comparisons and downstream scientific inference.

Federated Learning (FL) enables decentralized, privacy-preserving model training but struggles to balance global generalization and local personalization due to non-identical data distributions across clients. Personalized Fine-Tuning (PFT), a popular post-hoc solution, fine-tunes the final global model locally but often overfits to skewed client distributions or fails under domain shifts. We propose adapting Linear Probing followed by full Fine-Tuning (LP-FT), a principled centralized strategy for alleviating feature distortion (Kumar et al., 2022), to the FL setting. Through systematic evaluation across seven datasets and six PFT variants, we demonstrate LP-FT's superiority in balancing personalization and generalization. Our analysis uncovers federated feature distortion, a phenomenon where local fine-tuning destabilizes globally learned features, and theoretically characterizes how LP-FT mitigates this via phased parameter updates. We further establish conditions (e.g., partial feature overlap, covariate-concept shift) under which LP-FT outperforms standard fine-tuning, offering actionable guidelines for deploying robust personalization in FL.

In this paper, we study the finite-sample statistical rates of distributional temporal difference (TD) learning with linear function approximation. The purpose of distributional TD learning is to estimate the return distribution of a discounted Markov decision process for a given policy. Previous works on statistical analysis of distributional TD learning focus mainly on the tabular case. We first consider the linear function approximation setting and conduct a fine-grained analysis of the linear-categorical Bellman equation. Building on this analysis, we further incorporate variance reduction techniques in our new algorithms to establish tight sample complexity bounds independent of the support size $K$ when $K$ is large. Our theoretical results imply that, when employing distributional TD learning with linear function approximation, learning the full distribution of the return function from streaming data is no more difficult than learning its expectation. This work provide new insights into the statistical efficiency of distributional reinforcement learning algorithms.

The fundamental theorem of statistical learning states that binary PAC learning is governed by a single parameter – the Vapnik-Chervonenkis (VC) dimension – which determines both learnability and sample complexity. Extending this to multiclass classification has long been challenging, since Natarajan's work in the late 80s proposing the Natarajan dimension (Nat) as a natural analogue of VC. Daniely and Shalev-Shwartz (2014) introduced the DS dimension, later shown by Brukhim et al. (2022) to characterize multiclass learnability. Brukhim et al. also showed that Nat and DS can diverge arbitrarily, suggesting that multiclass learning is governed by DS rather than Nat. We show that agnostic multiclass PAC sample complexity is in fact governed by two distinct dimensions. Specifically, we prove nearly tight agnostic sample complexity bounds that, up to log factors, take the form $\frac{DS^{1.5}}{\epsilon} + \frac{Nat}{\epsilon^2}$ where $\epsilon$ is the excess risk. This bound is tight up to a $\sqrt{DS}$ factor in the first term, nearly matching known $Nat/\epsilon^2$ and $DS/\epsilon$ lower bounds. The first term reflects the DS-controlled regime, while the second shows that the Natarajan dimension still dictates asymptotic behavior for small $\epsilon$. Thus, unlike binary or online classification – where a single dimension (VC or Littlestone) controls both phenomena – multiclass learning inherently involves two structural parameters. Our technical approach departs from traditional agnostic learning methods based on uniform convergence or reductions to realizable cases. A key ingredient is a novel online procedure based on a self-adaptive multiplicative-weights algorithm performing a label-space reduction, which may be of independent interest.

Valuing residential property is inherently complex, requiring consideration of numerous environmental, economic, and property-specific factors. These complexities present significant challenges for automated valuation models (AVMs), which are increasingly used to provide objective assessments for property taxation and mortgage financing. The challenge of obtaining accurate and objective valuations for properties at a country level, and not just within major cities, is further compounded by the presence of multiple localised submarkets-spanning urban, suburban, and rural contexts-where property features contribute differently to value. Existing AVMs often struggle in such settings: traditional hedonic regression models lack the flexibility to capture spatial variation, while advanced machine learning approaches demand extensive datasets that are rarely available. In this article, we address these limitations by developing a robust statistical framework for property valuation in the Irish housing market. We segment the country into six submarkets encompassing cities, large towns, and rural areas, and employ a generalized additive model that captures non-linear effects of property characteristics while allowing feature contributions to vary across submarkets. Our approach outperforms both machine learning-based and traditional hedonic regression models, particularly in data-sparse regions. In out-of-sample validation, our model achieves R-squared values of 0.70, 0.84, and 0.83 for rural areas, towns, and Dublin, respectively, compared to 0.52, 0.71, and 0.82 from a random forest benchmark. Furthermore, the temporal dynamics of our model align closely with reported inflation figures for the study period, providing additional validation of its accuracy.

This paper proposes a novel framework to test for slope heterogeneity between time-varying coefficients in panel data models. Our test not only allows us to detect whether the coefficient functions are the same across all units or not, but also determines which of them are different and where these differences are located. We establish the asymptotic validity of our multiscale test. As an extension of the proposed procedure, we show how to use the results to uncover latent group structures in the model. We apply our methods to test for heterogeneity in the effect of U.S. monetary shocks on 49 foreign economies and itself. We find evidence that such heterogeneity indeed exists and we discuss the clustering results for two groups.

Low latency event-selection (trigger) algorithms are essential components of Large Hadron Collider (LHC) operation. Modern machine learning (ML) models have shown great offline performance as classifiers and could improve trigger performance, thereby improving downstream physics analyses. However, inference on such large models does not satisfy the $40\text{MHz}$ online latency constraint at the LHC. In this work, we propose PHAZE, a novel framework built on cryptographic techniques like hashing and zero-knowledge machine learning (zkML) to achieve low latency inference, via a certifiable, early-exit mechanism from an arbitrarily large baseline model. We lay the foundations for such a framework to achieve nanosecond-order latency and discuss its inherent advantages, such as built-in anomaly detection, within the scope of LHC triggers, as well as its potential to enable a dynamic low-level trigger in the future.

Stochastic multi-objective optimization (SMOOP) requires ranking multivariate distributions; yet, most empirical studies perform scalarization, which loses information and is unreliable. Based on the optimal transport theory, we introduce the center-outward q-dominance relation and prove it implies strong first-order stochastic dominance (FSD). Also, we develop an empirical test procedure based on q-dominance, and derive an explicit sample size threshold, $n^*(\delta)$, to control the Type I error. We verify the usefulness of our approach in two scenarios: (1) as a ranking method in hyperparameter tuning; (2) as a selection method in multi-objective optimization algorithms. For the former, we analyze the final stochastic Pareto sets of seven multi-objective hyperparameter tuners on the YAHPO-MO benchmark tasks with q-dominance, which allows us to compare these tuners when the expected hypervolume indicator (HVI, the most common performance metric) of the Pareto sets becomes indistinguishable. For the latter, we replace the mean value-based selection in the NSGA-II algorithm with $q$-dominance, which shows a superior convergence rate on noise-augmented ZDT benchmark problems. These results establish center-outward q-dominance as a principled, tractable foundation for seeking truly stochastically dominant solutions for SMOOPs.

Do nineteenth-century graphics still work for today's readers? To investigate this question, we conducted a controlled experiment evaluating three canonical historical visualizations- Nightingale's polar area diagram, Playfair's trade balance chart, and Minard's campaign map-against modern redesigns. Fifty-four participants completed structured question-answering tasks, allowing us to measure accuracy, response time, and perceived workload (NASA-TLX). We used mixed-effects regression models to find: Nightingale's diagram remained consistently effective across versions, achieving near-ceiling accuracy and low workload; Playfair's dual-axis redesign underperformed relative to both its historical and alternative versions; and Minard's map showed large accuracy gains under redesign but continued to impose high workload and long response times. These results demonstrate that some nineteenth-century designs remain effective, others degrade under certain modernizations, and some benefit from careful redesign. The findings indicate how perceptual encoding choices, task alignment, and cognitive load determine whether historical charts survive or require adaptation for contemporary use.

For decades, forensic statisticians have debated whether searching large DNA databases undermines the evidential value of a match. Modern surveillance faces an exponentially harder problem: screening populations across thousands of attributes using threshold rules rather than exact matching. Intuition suggests that requiring many coincidental matches should make false alerts astronomically unlikely. This intuition fails. Consider a system that monitors 1,000 attributes, each with a 0.5 percent innocent match rate. Matching 15 pre-specified attributes has probability \(10^{-35}\), one in 30 decillion, effectively impossible. But operational systems require no such specificity. They might flag anyone who matches any 15 of the 1,000. In a city of one million innocent people, this produces about 226 false alerts. A seemingly impossible event becomes all but guaranteed. This is not an implementation flaw but a mathematical consequence of high-dimensional screening. We identify fundamental probabilistic limits on screening reliability. Systems undergo sharp transitions from reliable to unreliable with small increases in data scale, a fragility worsened by data growth and correlations. As data accumulate and correlation collapses effective dimensionality, systems enter regimes where alerts lose evidential value even when individual coincidences remain vanishingly rare. This framework reframes the DNA database controversy as a shift between operational regimes. Unequal surveillance exposures magnify failure, making “structural bias” mathematically inevitable. These limits are structural: beyond a critical scale, failure cannot be prevented through threshold adjustment or algorithmic refinement.

In this paper, we study transfer learning for high-dimensional factor-augmented sparse linear models, motivated by applications in economics and finance where strongly correlated predictors and latent factor structures pose major challenges for reliable estimation. Our framework simultaneously mitigates the impact of high correlation and removes the additional contributions of latent factors, thereby reducing potential model misspecification in conventional linear modeling. In such settings, the target dataset is often limited, but multiple heterogeneous auxiliary sources may provide additional information. We develop transfer learning procedures that effectively leverage these auxiliary datasets to improve estimation accuracy, and establish non-asymptotic $\ell_1$- and $\ell_2$-error bounds for the proposed estimators. To prevent negative transfer, we introduce a data-driven source detection algorithm capable of identifying informative auxiliary datasets and prove its consistency. In addition, we provide a hypothesis testing framework for assessing the adequacy of the factor model, together with a procedure for constructing simultaneous confidence intervals for the regression coefficients of interest. Numerical studies demonstrate that our methods achieve substantial gains in estimation accuracy and remain robust under heterogeneity across datasets. Overall, our framework offers a theoretical foundation and a practically scalable solution for incorporating heterogeneous auxiliary information in settings with highly correlated features and latent factor structures.

This work addresses a key challenge in inventory management by developing a stochastic model that describes the dynamic distribution of inventory stock over time without assuming a specific demand distribution. Our model provides a flexible and applicable solution for situations with limited historical data and short-term predictions, making it well-suited for the Newsvendor problem. We evaluate our model's performance using real-world data from a large electronic marketplace, demonstrating its effectiveness in a practical forecasting scenario.

Most estimators collapse all uncertainty modes into a single confidence score, preventing reliable reasoning about when to allocate more compute or adjust inference. We introduce Uncertainty-Guided Inference-Time Selection, a lightweight inference time framework that disentangles aleatoric (data-driven) and epistemic (model-driven) uncertainty directly in deep feature space. Aleatoric uncertainty is estimated using a regularized global density model, while epistemic uncertainty is formed from three complementary components that capture local support deficiency, manifold spectral collapse, and cross-layer feature inconsistency. These components are empirically orthogonal and require no sampling, no ensembling, and no additional forward passes. We integrate the decomposed uncertainty into a distribution free conformal calibration procedure that yields significantly tighter prediction intervals at matched coverage. Using these components for uncertainty guided adaptive model selection reduces compute by approximately 60 percent on MOT17 with negligible accuracy loss, enabling practical self regulating visual inference. Additionally, our ablation results show that the proposed orthogonal uncertainty decomposition consistently yields higher computational savings across all MOT17 sequences, improving margins by 13.6 percentage points over the total-uncertainty baseline.

Multivariable Mendelian Randomization (MVMR) estimates the direct causal effects of multiple risk factors on an outcome using genetic variants as instruments. The growing availability of summary-level genetic data has created opportunities to apply MVMR in high-dimensional settings with many strongly correlated candidate risk factors. However, existing methods face three major limitations: weak instrument bias, limited interpretability, and the absence of valid post-selection inference. Here we introduce MVMR-PACS, a method that identifies signal-groups – sets of causal risk factors with high genetic correlation or indistinguishable causal effects – and estimates the direct effect of each group. MVMR-PACS minimizes a debiased objective function that reduces weak instrument bias while yielding interpretable estimates with theoretical guarantees for variable selection. We adapt a data-thinning strategy to summary-data MVMR to enable valid post-selection inference. In simulations, MVMR-PACS outperforms existing approaches in both estimation accuracy and variable selection. When applied to 27 lipoprotein subfraction traits and coronary artery disease risk, MVMR-PACS identifies biologically meaningful and robust signal-groups with interpretable direct causal effects.

In many settings, a data curator links records from two files to produce datasets that are shared with secondary analysts. Analysts use the linked files to estimate models of interest, such as regressions. Such two-stage approaches do not necessarily account for uncertainty in model parameters that results from uncertainty in the linkages. Further, they do not leverage the relationships among the study variables in the two files to help determine the linkages. We propose a multiple imputation framework to address these shortcomings. First, we use a bipartite Bayesian record linkage model to generate multiple plausible linked datasets, disregarding the information in the study variables. Second, we presume each linked file has a mixture of true links and false links. We estimate the mixture model using information from the study variables. Through simulation studies under a regression setting, we demonstrate that estimates of the regression model parameters can be more accurate than those based on an analogous two-stage approach. We illustrate the integrated approach using data from the Survey on Household Income and Wealth, examining a regression involving the persistence of income.

Learning causality from observational data has received increasing interest across various scientific fields. However, most existing methods assume the absence of latent confounders and restrict the underlying causal graph to be acyclic, assumptions that are often violated in many real-world applications. In this paper, we address these challenges by proposing a novel framework for causal discovery that accommodates both cycles and latent confounders. By leveraging the identifiability results from noisy independent component analysis and recent advances in factor analysis, we establish the unique causal identifiability under mild conditions. Building on this foundation, we further develop a fully Bayesian approach for causal structure learning, called BayCausal, and evaluate its identifiability, utility, and superior performance against state-of-the-art alternatives through extensive simulation studies. Application to a dataset from the Women's Interagency HIV Study yields interpretable and clinically meaningful insights. To facilitate broader applications, we have implemented BayCausal in an R package, BayCausal, which is the first publicly available software capable of achieving unique causal identification in the presence of both cycles and latent confounders.

Self-consistency (SC) is a widely used test-time inference technique for improving performance in chain-of-thought reasoning. It consists of generating multiple responses, or “samples", from a large language model (LLM) and selecting the most frequent answer. This procedure can naturally be viewed as a majority vote or empirical mode estimation. Despite its effectiveness, self-consistency is prohibitively expensive at scale when naively applied to datasets, and it lacks a unified theoretical understanding of sample efficiency and scaling behavior. In this paper, we provide the first comprehensive analysis of SC's scaling behavior and its variants, drawing on mode estimation and voting theory. We derive and empirically validate power law scaling for self-consistency across datasets, and analyze the sample efficiency for fixed-allocation and dynamic-allocation sampling schemes. From these insights, we introduce Blend-ASC, a novel variant of self-consistency that dynamically allocates samples to questions during inference, achieving state-of-the-art sample efficiency. Our approach uses 4.8 times fewer samples than vanilla SC on average, outperforming both fixed- and dynamic-allocation SC baselines, thereby demonstrating the superiority of our approach in terms of efficiency. In contrast to existing variants, we note that Blend-ASC is hyperparameter-free, supports batching, and can fit any budget of samples, ensuring it can be easily applied to any self-consistency application.

High-dimensional data often contain low-dimensional signals obscured by structured background noise, which limits the effectiveness of standard PCA. Motivated by contrastive learning, we address the problem of recovering shared signal subspaces from positive pairs, paired observations sharing the same signal but differing in background. Our baseline, PCA+, uses alignment-only contrastive learning and succeeds when background variation is mild, but fails under strong noise or high-dimensional regimes. To address this, we introduce PCA++, a hard uniformity-constrained contrastive PCA that enforces identity covariance on projected features. PCA++ has a closed-form solution via a generalized eigenproblem, remains stable in high dimensions, and provably regularizes against background interference. We provide exact high-dimensional asymptotics in both fixed-aspect-ratio and growing-spike regimes, showing uniformity's role in robust signal recovery. Empirically, PCA++ outperforms standard PCA and alignment-only PCA+ on simulations, corrupted-MNIST, and single-cell transcriptomics, reliably recovering condition-invariant structure. More broadly, we clarify uniformity's role in contrastive learning, showing that explicit feature dispersion defends against structured noise and enhances robustness.

Incomplete multi-view unsupervised feature selection (IMUFS), which aims to identify representative features from unlabeled multi-view data containing missing values, has received growing attention in recent years. Despite their promising performance, existing methods face three key challenges: 1) by focusing solely on the view-missing problem, they are not well-suited to the more prevalent mixed-missing scenario in practice, where some samples lack entire views or only partial features within views; 2) insufficient utilization of consistency and diversity across views limits the effectiveness of feature selection; and 3) the lack of theoretical analysis makes it unclear how feature selection and data imputation interact during the joint learning process. Being aware of these, we propose CLIM-FS, a novel IMUFS method designed to address the mixed-missing problem. Specifically, we integrate the imputation of both missing views and variables into a feature selection model based on nonnegative orthogonal matrix factorization, enabling the joint learning of feature selection and adaptive data imputation. Furthermore, we fully leverage consensus cluster structure and cross-view local geometrical structure to enhance the synergistic learning process. We also provide a theoretical analysis to clarify the underlying collaborative mechanism of CLIM-FS. Experimental results on eight real-world multi-view datasets demonstrate that CLIM-FS outperforms state-of-the-art methods.

This paper proposes a novel Bayesian framework for solving Poisson inverse problems by devising a Monte Carlo sampling algorithm which accounts for the underlying non-Euclidean geometry. To address the challenges posed by the Poisson likelihood – such as non-Lipschitz gradients and positivity constraints – we derive a Bayesian model which leverages exact and asymptotically exact data augmentations. In particular, the augmented model incorporates two sets of splitting variables both derived through a Bregman divergence based on the Burg entropy. Interestingly the resulting augmented posterior distribution is characterized by conditional distributions which benefit from natural conjugacy properties and preserve the intrinsic geometry of the latent and splitting variables. This allows for efficient sampling via Gibbs steps, which can be performed explicitly for all conditionals, except the one incorporating the regularization potential. For this latter, we resort to a Hessian Riemannian Langevin Monte Carlo (HRLMC) algorithm which is well suited to handle priors with explicit or easily computable score functions. By operating on a mirror manifold, this Langevin step ensures that the sampling satisfies the positivity constraints and more accurately reflects the underlying problem structure. Performance results obtained on denoising, deblurring, and positron emission tomography (PET) experiments demonstrate that the method achieves competitive performance in terms of reconstruction quality compared to optimization- and sampling-based approaches.

The identification of domain sets whose outcomes belong to predefined subsets can address fundamental risk assessment challenges in climatology and medicine. Existing approaches for inverse domain estimates require restrictive assumptions, including domain density and continuity of function near thresholds, and large-sample guarantees, which limit the applicability. Besides, the estimation and coverage depend on setting a fixed threshold level, which is difficult to determine. Recently, Ren et al. (2024) proved that confidence sets of multiple levels can be simultaneously constructed with the desired confidence non-asymptotically through inverting simultaneous confidence bands. Here, we present the SCoRES R package, which implements Ren's approach for both the estimation of the inverse region and the corresponding simultaneous outer and inner confidence regions, along with visualization tools. Besides, the package also provides functions that help construct SCBs for regression data, functional data and geographical data. To illustrate its broad applicability, we present three rigorous examples that demonstrate the SCoRES workflow.