Papers

Learning distributions of longitudinal data is central to tasks such as visualization, completion, classification, and synthetic data generation, but it remains statistically challenging because longitudinal observations are often irregular, sparse, and collected from only a limited number of subjects. To address this, we develop a novel generative framework, termed panel flow matching (PFM), for learning longitudinal distributions by pooling information across time via a continuous panel flow model. PFM combines a forward flow-matching step with a backward kernel-fitting step, yielding a flexible and data-adaptive approach for capturing complex distributional structures. We apply PFM to estimate panel densities, namely the cross-sectional densities of longitudinal data, and establish statistical guarantees under irregular and sparse sampling designs. Under this, PFM naturally supports tasks including longitudinal completion, synthetic data generation, and classification, without requiring a preliminary dimension-reduction step to handle data irregularity. Extensive simulations demonstrate that PFM outperforms existing methods across these tasks. We further apply PFM to a vaginal microbiome longitudinal dataset from 188 pregnancies labeled as term or preterm, where it improves classification accuracy and reveals time-varying distributional differences between the two groups.

This paper introduces the R package spca, which provides a computational framework for least squares sparse principal component analysis (LS-SPCA). Unlike other SPCA methods, LS-SPCA generates uncorrelated sparse principal components (sPCs) that effectively maximize the explained variance while maintaining strong correlations with standard principal components (PCs). The framework also includes more computationally efficient variants that produce mildly correlated sPCs, which often have lower cardinality while explaining equal or greater variance than the LS-SPCA optimal sPCs. The spca package is built on an efficient C++ backend for matrix computations, with distinct engines for tall and fat matrices, and a flexible R frontend. The user interface offers several options for computing sPCs, such as deciding whether sparsification should stop when a threshold for cumulative variance explained or R2 with the PCs is reached, and choosing between simple forward selection, stepwise forward selection, or backward elimination for variable selection. In addition to the print(), summary(), and plot() methods, the package includes tools for comparing different "spca" solutions, grouping sparse loadings, and representing foreign SPCA solutions as "spca" objects. This article demonstrates with real datasets the use of the package in a typical LS-SPCA workflow and briefly contrasts LS-SPCA with conventional SPCA solutions . Then it compares different LS-SPCA solutions obtained from the dataset. Finally, the performance of spca on large tall and fat matrices is discussed, showing that spca offers a computationally efficient alternative for computing interpretable sPCs.

A growing number of techniques leverage the spatial structures that underlie many real-world datasets. Despite these advances, the complementary task of estimating spatial structures and understanding their role within these techniques has often been overlooked. In neurophysiological data analysis specifically, numerous methods exist to estimate brain connectivity, but most are not explicitly model-based, dynamic, multivariate, or directed. To address these limitations, we previously introduced noise-driven heat modelling on graphs for neurophysiological connectivity estimation. In this study, we extend this framework by relaxing earlier noise assumptions and adding regularisation to improve robustness. We also develop a simulation procedure to characterise and evaluate our technique in a controlled setting. Finally, we demonstrate that the technique is able to capture meaningful spatial structure across two experiments, each using two real-world datasets. The explicit model formulation of our connectivity estimator has the potential to improve the interpretability of graph-based techniques across a wide range of applications. The code implementing our method is available at https://github.com/sgoerttler/Heat_Connectivity.

Principal stratification provides a foundational framework for causal inference with intermediate outcomes by defining causal effects within subpopulations, yet existing work has largely focused on average effects across strata rather than treatment effect heterogeneity within strata. Such within-stratum heterogeneity informs individualized treatment decisions but the associated methods are sparse. We address this gap by studying the identification and estimation of the conditional principal causal effects under principal ignorability combined with an odds ratio sensitivity parameterization, which relaxes the monotonicity assumption. To efficiently learn these estimands, we propose a novel doubly cross-fit doubly robust machine learner that resolves the nested nuisance structure inherent to principal stratification. Leveraging sequential orthogonal debiased machine learning with regularized least-squares sieves, we derive $\mathcal{L}^2$ and uniform limit theory, establish oracle efficiency, and construct uniform confidence bands for the proposed estimator. We use simulations to demonstrate the finite-sample performance of our estimator, and provide an empirical analysis of a randomized trial in acute lung injury, revealing informative patterns of treatment effect heterogeneity within the always-survivor subpopulation.

Standard statistical methods are often inadequate for modeling the joint dependence between linear and circular variables, and existing methods for modeling this dependence are designed only for continuous variables. However, circular data are frequently observed on a finite set of equally spaced directions, either due to rounding prior to reporting or because of the experimental design employed for data collection. To address this gap, we propose a flexible, analytically tractable model for jointly representing a discrete circular and a continuous linear variable. The construction combines a wrapped symmetric geometric distribution, a Weibull distribution, and a trigonometric linking function. This formulation yields closed-form expressions for the joint, marginal, and conditional distributions. The choice of the Weibull distribution facilitates direct sample generation using the inverse transform technique. Additionally, it provides explicit expressions for conditional moments, enabling a flexible circular-linear regression framework. We detail the theoretical interpretation of the model parameters, mathematically establishing the monotonicity of the conditional mean and variance with respect to the dependence parameters. The performance of the estimators is demonstrated through extensive simulations, and the utility of the model is illustrated by analyzing two empirical environmental datasets.

We introduce a novel goodness-of-fit (GOF) procedure based on Beta-tree partitions. A Beta-tree produces a data-adaptive partition of the sample space into regions and provides guaranteed finite sample confidence intervals for the probability contents of each region. The proposed test assesses whether the probabilities assigned by a null distribution $F_0$ fall within these intervals, thereby quantifying agreement between the model and the data. A key application is the selection of the number of components in a mixture model, where the null distribution is constructed via $k$-means clustering. In contrast to classical global GOF tests such as Kolmogorov-Smirnov or Anderson-Darling, which quantify the discrepancy through a single global statistic, our method is designed to detect local departures from the null and to identify regions of model misspecification. We demonstrate the efficiency of our test in detecting departures from the null on some simulated and real datasets.

We show that net demand for liquidity by algo strategies is identifiable from its trade and price history alone, with no knowledge of its signal or optimization problem. An exact multi-period regret decomposition implies that the sign of this statistic classifies a linear strategy as a net liquidity consumer or provider, recovering the Kyle (1985) informed-trader/market-maker dichotomy from observables alone. Under an AR(1) cost process, the same statistic equals the product of strategy size and the squared Roll (1984) implied spread, making the correction a direct proxy for prevailing illiquidity. Extending to endogenous price impact and aggregating across N correlated strategies yields a liquidity-balance condition whose violation produces welfare loss scaling as N squared, a closed-form fire-sale externality. We calibrate to CRSP equity data (2016-2025), tracking implied spreads through the COVID-19 and 2022 rate-shock episodes, with an estimator computable in O(Tnd) time.

Time-varying treatment effects, surrogate-identified treatment effects, and mediation effects can all be written as recursive regressions, in which each regression's predicted values become generated outcomes for the next regression. We study how standard causal estimators behave in this setting. Formally, we compare the recursive plug-in, recursive balancing weight, and recursive doubly robust estimators. When every stage is fitted by ordinary least squares (OLS), the three recursive estimators coincide in any finite sample, whether or not the models are correctly specified. As such, estimation by recursively regressing generated outcomes is numerically equivalent to estimation by recursively balancing generated regressors. Under ridge penalisation for the balancing weights, the doubly robust estimator is a backward recursion of stage-wise blends of penalised and OLS regressions. The weight on the recursive OLS regression decays geometrically in the number of time periods. Therefore, the intuition from the cross-sectional setting, where the bias correction moves the estimator towards OLS, applies less and less as the number of time periods increases. For general convex penalties, we derive an identity at each stage.

When datasets contain outliers, robust regression is a well-established alternative to Ordinary Least Squares. A commonly employed robust estimator is Least Trimmed Squares (LTS), which computes the regression coefficients from a subset of observations. Determining the exact solution corresponds to a combinatorial problem with prohibitive computational costs, even for instances of moderate dimension. Thus, the most prevalent approach in practice remains a heuristic known as Fast-LTS. Although the heuristic often performs effectively, certain elements of the approach remain open to improvement. In particular, its core procedure provides robust results only when initialized with a large number of starting points. To address the heuristic's limitations, this paper reformulates the LTS problem as a concave minimization problem subject to a capped simplex constraint, and proposes the successive Boosted Difference of Convex Functions Algorithm (sBDCA) as a solution method. Theoretically, we establish via the Łojasiewicz property that sBDCA converges to a local solution with a linear rate in the fastest case. To ensure robustness from a single initialization in practice, we derive and integrate a problem-specific preconditioning matrix into the algorithmic setup. Building on this theoretical foundation, we conduct numerical studies on various synthetic and real-world datasets to demonstrate the effectiveness of sBDCA with preconditioning. Specifically, we show that our approach is up to 3.25 times faster than Fast-LTS and achieves up to 90% lower objective function values, particularly in high-dimensional settings. As all code is openly available, this paper further provides a practical guide to robust regression in Python.

Physicians now pose millions of clinical questions to AI tools each week, yet these tools are evaluated largely on hypothetical or exam-style questions, not those actually asked in practice. We report a blinded evaluation built on 620 Real-world Point-Of-Care Queries (Real-POCQi) submitted to the OpenEvidence (OE) platform by physicians spanning 30 specialties, as well as 187 questions from HealthBench. 149 practicing physicians across 36 states made head-to-head comparisons between answers from three frontier general-purpose models (Claude Opus 4.8, Gemini 3.1 Pro, and GPT-5.5) and a specialized clinical tool (OE), with graders matched to each question's specialty. When comparing answers along five dimensions relevant to clinical decision support – accuracy, clinical utility, source quality, verifiability, & completeness – physicians scored the specialized tool highest on all axes; in the primary analysis on Real-POCQi, win differences (margins between win and loss rates) ranged from 25 to 39 percentage points (p<0.001). Results remained consistent in sensitivity analyses stratifying by citation display, answer length, OE-user status, and Real-POCQi versus HealthBench. In parallel, LLM judges were found to systematically differ from expert judges, though both generally agreed on the best model. These findings underscore two conclusions: (i) AI tool evaluations should reflect real-world query distributions and use expert judges that mirror the specialization defining modern medicine and (ii) the consistent advantage of the specialized tool over general-purpose models does not necessarily mean that the latter cannot serve similar purposes, but that targeted engineering and customization can yield meaningful gains in performance for its users. We release Real-POCQi as a public benchmark, as well as the prespecified statistical analysis for reproducing results of this study.

Predicting the aerodynamic performance (e.g. lift, drag, and moment coefficients) of an aircraft is challenging – computational models are biased and direct simulations are prohibitive. A pragmatic way to overcome this limitation is by calibrating low-fidelity computational predictions with experimental measurements. This, however, requires calibrating against sparse measurements contaminated with uncertainty in both the control inputs and the measured aerodynamic response. We develop a methodology to address this problem based on Gaussian process surrogates and the classical Kennedy-O'Hagan calibration. A surrogate model learned on abundant-but-cheap low-fidelity data is calibrated with a sparse set of measurement data. Crucialy, we develop a Bayesian latent Gaussian process based approach that marginalizes the calibrated surrogate model over the input uncertainty, while also matching the marginal mean and variance of the measured output uncertainty. Once calibrated, our surrogate model predicts the uncertainty in aerodynamic coefficients with very high accuracy, including at extrapolative input settings. We validate our calibrated surrogate model predictions against measurement data with true uncertainty intervals to demonstrate that the model places $94.2-95.8\%$ of its predictive samples inside the released $95\%$ truth intervals, with endpoint cumulative probabilities very close to the nominal 0.025 and 0.975 levels.

The common factor analytic model is related to Helmholtz and Boltzmann machines, can be conceived as a linear autoencoder, or can be thought of as a single-hidden-layer generative neural network. We thus consider it a basal generative representation learner that can be used as a minimal model for studying the foundational characteristics of (deep) generative model architectures. We focus on the fundamental problem of indeterminacy in latent factor projections. This indeterminacy implies that, even when the intrinsic dimension of the latent vector is known, regularity conditions are met, and rotational indeterminacy is resolved, an inherent indefiniteness in the retrieval of causative latent sources remains: they will be uncertain, distributionally deviant, and non-unique. This can have major implications for data representation but remains an elusive issue, even to practitioners and theorists well-versed in the factor model. Moreover, this classic psychometric problem is intricately related to the modern issue of latent variable collapse in the variational autoencoder framework for deep generative modeling. Here, we assess this indeterminacy from various perspectives and show how these are mathematically and conceptually related and we discuss subsequent implications for the Psychometrics, Statistics, and Artificial Intelligence communities. We show that one has latent factor determinacy across all its facets when the feature-dimension grows to infinity. This feeds into an essentially distribution-free estimation approach in the sample case when the number of features grows very large. We conclude, as these are emergent properties at scale, that the factor model is suited for representation learning of very-high-dimensional data.

Consider an analyst interested in predicting the size of an effect. She has identified a set of prior published studies of similar effects. We provide a toolkit for (i) summarizing the prior literature, (ii) making predictions of effects in new contexts, and (iii) correcting for the bias from selectivity in the prior literature. We illustrate these methods with empirical examples from labor, public, behavioral, environmental, and development economics. Some of the tools are relevant even when only three prior studies are available. We show how it is possible to use covariates to transparently make predictions for a new context by reweighting prior estimates. The mean effect 0 after correcting for selectivity - is between 12% and 21% of the simple mean in our empirical examples. We conclude with a cookbook for practitioners producing meta-analyses.

Biomedical research is increasingly relying on readily available routine data, such as electronic health records. Routinely collected data, as well as datasets from large cohorts, are often prone to measurement error which, if not addressed in analyses, can bias study results and ultimately mislead clinical decision-making and potentially harm patients. For this setting, methods that address errors in the outcome and multiple covariates are needed. In this tutorial, we will review available methods to address for errors in both outcomes and covariates. We will illustrate methods with use of a running example in order to compare the methods directly. Both the data and analytic code are provided for the user so that they may easily reproduce results in each example. We conclude the tutorial with a discussion of the different approaches and highlight areas of future work needed for this setting.

We study the leading-order fluctuation of stochastic gradient Euler-Maruyama estimators for generalized non-reversible Langevin dynamics. Under structural assumptions tailored to the small-stepsize central limit theorem and under an unbiased stochastic gradient oracle, we prove that the empirical average over a horizon of order the inverse squared stepsize satisfies a central limit theorem in the vanishing-stepsize regime. The limiting variance is characterized through the Poisson equation of the limiting full-gradient diffusion. We then rewrite this constant in an operator form that links it to the continuous-time asymptotic variance and, under standard operator-theoretic assumptions, derive a sufficient condition under which an anti-symmetric perturbation strictly reduces the leading-order fluctuation constant relative to the reversible baseline. We also identify bounded smooth predictive observables that re directly covered by the main theorem. As a separate Gaussian calculation beyond the bounded-test-function regime, we obtain closed-form formulas for quadratic Hamiltonians and linear observables. The framework covers non-reversible Langevin dynamics and augmented-state examples including Hessian-free high-resolution dynamics and a positive-definite subclass of gradient-adjusted underdamped Langevin dynamics that allow stochastic gradients. Numerical experiments on basic examples and Bayesian linear regression using synthetic data, and Bayesian logistic regression using real data support the predicted Gaussian fluctuations and show that the non-reversible schemes consistently reduce the root mean squared error (RMSE) relative to their reversible baselines.

The main objective of this paper is to propose a general framework for prediction based on different sources of multimodal data in the healthcare domain. We evaluated whether frozen medical large language model (LLM) representations can serve as a shared embedding space for multimodal primary diagnosis category prediction. We propose a backbone-flexible probing framework that serializes structured electronic health record (EHR) variables and leakage-pruned discharge-note sections into a shared frozen-LLM representation space and recovers primary diagnosis categories with linear probes, without fine-tuning the LLM. On a MIMIC-IV cohort of 13,645 admissions (seven diagnosis categories), we probed inputs of different modalities (Structured-only, Unstructured-only, and Combined) across transformer layers and six frozen backbones, compared against XGBoost and three text baselines (CAML, LAAT, PLM-ICD) under a matched split and context budget, and tested robustness across datasets on MIMIC-III. The Combined terminal-layer probe reached 88.86% top-1 accuracy on MIMIC-IV, significantly outperforming CAML and PLM-ICD and statistically comparable to LAAT; Combined accuracy ranged from 87.69% to 89.25% across six backbones. On MIMIC-III, in-domain Combined probing reached 74.18% accuracy and significantly exceeded all three baselines, and in a transfer-learning test a 2.8M-parameter adapter on frozen MIMIC-IV representations reached 92.51% accuracy, exceeding adapted LAAT (85.54%). Frozen LLM embeddings unify structured and narrative EHR information: the Combined configuration was strongest in every setting, and its representations transferred across datasets through a compact adapter. Multimodal probing of frozen LLM representations provides a practical, backbone-flexible approach for studying EHR modalities and adapting clinical representations across datasets.

Machine Learning (ML) algorithms, such as k-Nearest Neighbours (kNN) or random forest, eschew the ideal of true data models in favour of predictive performance. However, minimising the MSE or F-score cannot lead to unbiasedness directly, which is important in many situations such as official statistics. We study the conditions of algorithmic ML, other than the existence and knowledge of true data models, which lead to unbiased prediction or classification for a given finite population, including how the training data may be sampled from the population, how a trained prediction algorithm can be tuned to achieve unbiased prediction or classification for that population, and how the performance of out-of-sample prediction or classification can be assessed unbiasedly. The inference is based on the known probability design of samples and training sets, rather than any assumed distributions or models.

A critical assumption of observational studies is that all confounding variables must be known and sufficiently adjusted for to estimate causal effects. An implicit, and often overlooked, aspect of this assumption is that all confounding variables have been measured without error. In the social and medical sciences, latent traits such as motivation, self-efficacy, and ability measures are likely confounding variables. Because latent traits are not directly observable, conventional approaches to adjust for them in observational studies rely on collecting responses to individual items on a test or survey instrument and then adjust for sum scores, measurement model-derived ability estimates, or item responses directly. Through a process we describe as measurement induced confounding, we show that measurement error propagates through the estimation process and that current conventional approaches to adjusting for latent traits in observational studies produce biased estimates of the average treatment effect with incorrectly calibrated coverage properties. A critical implication of this finding is that current observational studies that attempt to adjust for latent confounding variables likely put forth biased causal estimates with incorrect uncertainty intervals. We show that measurement induced confounding can be resolved through a Bayesian Joint Estimation approach that simultaneously estimates the measurement model, the treatment assignment model, and the response model.

This paper presents a methodological framework for estimating the comprehensive cohort causal effect (CCCE) in mixed-design clinical studies that combine randomized controlled trials (RCTs) and parallel observational study (OBS). Our approach is designed to evaluate robustness against unmeasured confounding in the OBS arm and to handle outcomes that are missing at random in either the RCT or OBS arm. By employing a semiparametric theory-based sensitivity analysis framework, we derive the efficient influence function for the CCCE, parameterized by sensitivity parameters. We propose a one-step bias-corrected estimator that allows for flexible modeling and establish conditions under which our CCCE estimator is $\sqrt{n}$-consistent. To illustrate our methods, we apply them to the TOIB study, which evaluates the efficacy and safety of oral versus topical ibuprofen in managing chronic knee pain among older adults. We also evaluate the performance of the proposed methodology in a realistic simulation study.

[Working Draft] Compositional data are central to microbial, ecological, and environmental research, yet often have four features that are difficult to accommodate jointly: exact zeros, latent dependence among components, high-dimensionality, and a unit-sum constraint that induces a non-Euclidean geometry. Conventional Dirichlet-type and logistic-normal models address these features only partially. Projected Gaussian models offer a directional representation that captures exact zeros and latent dependence; however, support correctness on the simplex requires either truncation or folding, both of which become computationally prohibitive as the dimension grows. We develop an Active-set Ray-based Compositional (ARC) framework, which retains the benefits of projected Gaussian models while remaining computationally feasible in high-dimensional settings. In this framework, we map compositions to the nonnegative orthant of the unit hypersphere and specify an active-set process that governs which components are present. Conditional on the active set, the positive subcomposition is modeled by evaluating a latent Gaussian density along positive rays of the active subspace with the radius treated as an auxiliary variable. Such a construction (i) separates the active-set process that governs which components are present from the positive subcomposition on the active components, (ii) preserves a latent Gaussian interpretation, and (iii) accommodates arbitrary latent dependence. Thus, the framework is conducive to high-dimensional applications in which exact zeros and shared positive responses are scientifically central. Conceptually, the proposed framework reframes a composition as an observed direction of a latent abundance vector with an unobserved magnitude and an explicitly modeled active set.

The theory of full conformal prediction uses deterministic non-conformity measure, but modern usage of full conformal prediction often relies on machine learning training, making stochasticity inevitable. A simple sufficient condition of almost sure permutation invariance of the non-conformity measure can be too restrictive, so many have suggested the relaxation to permutation in distribution as a condition for full conformal prediction validity. We, however, show that this commonly known condition is actually insufficient. We then provide a correct sufficient condition: Conditional Independence & Permutation Invariance in Distribution, which encompasses several stochastic settings that may be used in machine learning.

We present a justification of the use of Inverse Probability Weighting (IPW) in a post-Bayesian framework, in which the bias-correction provided by IPW in a frequentist context is reframed as a reweighting of the Kullback-Leibler (KL) divergence between the statistical model and the true data-generating parameter value. We provide a coherent argument in support of this approach, including theoretical results concerning convergence and properties of the generalised belief posteriors. We present examples demonstrating the utility of post-Bayesian IPW in practice: these include two simulated examples of inference under selection bias in the observed data, and a large-scale real-data example concerning systematic biases present in registry data when using prostate-specific antigen (PSA) to predict prostate cancer mortality. The empirical and theoretical results together show the utility of IPW to address classes of problems previously intractable within a Bayesian approach.

Large Language Models (LLMs) often face ethical tradeoffs in which several responses may be defensible but express different priorities, such as fairness, honesty, courage, or restraint. We introduce VirtueMap, a framework for describing these patterns through an Aristotelian virtue-ethics lens. Instead of asking for a single correct answer, VirtueMap asks humans or LLMs to rank all five responses to each of seven general, non-lethal, non-political, and non-religious ethical dilemmas. To define the reference orderings used for scoring, we first proposed, for each dilemma and virtue, an ordering of the five responses from most to least expressive of that virtue. We then collected more than 100 respondent evaluations per ordering and retained it as operational ground truth only when at least 95% confirmed it. Rankings are scored against these retained orderings using normalized Borda alignment, yielding profiles over Practical Wisdom, Justice, Truthfulness, Courage, and Temperance. We apply VirtueMap to nine LLM families in a repeated-run evaluation and find high mean rank consistency (90.3%), with the largest differences appearing on Courage, Temperance, and Justice. We also release an interactive website that computes profiles locally in the browser and compares respondents with measured LLM profiles.

In this paper, we extend Stein's method to the symmetric matrix normal distribution. In particular, we obtain a Stein characterization of the symmetric matrix normal distribution involving the extended generator of the symmetric matrix Ornstein-Uhlenbeck process, present a semigroup representation of the solution of the corresponding Stein equation, and establish regularity estimates for the solution. This framework of Stein's method for symmetric matrix normal approximation complements the recent theory of Stein's method for matrix normal approximation, and we make an explicit connection between these frameworks. We apply this theory to derive a Wasserstein distance bound for the symmetric matrix normal approximation of the Wishart distribution.

People overthink; language models over-sample, and the extra effort can talk both into a worse answer. Reasoning systems answer a hard question by sampling it many times (test-time scaling), and the more they draw, the more often a correct answer turns up somewhere, so coverage, the fraction of problems with at least one correct try, climbs and appears to be progress. But a deployed system must return one answer, and choosing it, not knowing which try is right, is selection; selection is capped, and past a point extra samples only make the model surer of a confident mistake, even as every draw adds cost. The gap between climbing coverage and stalled selection, the identifiability gap, is the answer a model can produce but not pick. So the real question is not whether to sample but how far, and the answer is: not far. For picking an answer, the vote has already settled within a few dozen draws, the modal ceiling; for scoring a benchmark, sooner still, the correlation ceiling. Beyond that, extra draws cost compute and add nothing, and can even make the answer worse. This paper turns the cutoff into a single number, the effective number of samples, that any sampling run already reveals. The bottleneck is recognizing a right answer, not generating one.

AI judges offer a scalable, low-cost alternative to human evaluation, but their outputs can be biased relative to human preferences and highly item-dependent, varying across judges, tasks, and domains. When uncalibrated AI evaluations are used for model ranking, item scoring, or population-level quality reporting, these biases can directly distort downstream decisions. We propose BACON, a four-stage pipeline that combines budgeted human calibration with multiple AI-judge outputs to produce more accurate annotations. BACON constructs full-coverage auxiliary features for every item, including multi-judge scores, token-level uncertainty statistics, and contextual embeddings. It then collects human labels for a small sampled subset and trains a cross-fitted outcome model to generate calibrated item-level surrogate predictions. These predictions support two use cases: population-level estimation of summary metrics, such as means or quantiles, using an augmented estimating-equation estimator with valid confidence intervals; and individual-level surrogate scoring for item ranking and annotation. BACON treats AI judges as auxiliary measurements rather than ground truth: human labels provide the calibration anchor, while AI-derived signals improve efficiency. Across diverse tasks, domains, and labeling budgets, BACON improves predictive accuracy and ranking consistency, and reduces bias and variance relative to raw AI outputs and purely human-label-based methods. These results show that BACON offers a practical, statistically grounded framework for scalable evaluation with limited human annotation.

Bayesian and multiplicative-weights updates reweight experts, models, or actions from sequential feedback. We show that the regret of any such update obeys an exact information-accounting identity. On each round, the learner's excess loss to any chosen comparator is the sum of an immediate cost for the uncertainty exposed by the round and a reduction in the information distance from the learner's current weights to the comparator. The cumulative cost defines a pathwise uncertainty clock, the intrinsic time of the realized sequence. Summing one-step balances yields two exact adaptive decompositions of cumulative regret, one for each natural way of composing the update across rounds. Because the decompositions are exact, favorable stochastic or low-noise regimes appear as self-bounding properties of the realized intrinsic time. The accounting also fixes a learning rate, inverse in the square root of intrinsic time. That schedule is competitive with adaptive baselines in selected online-learning settings. The same calculus covers Hedge, optimistic and side-information variants, continuous priors, boosting, online convex optimization, contextual bandits, and repeated games: the pathwise account is the same in every case.

Value-of-information (VOI) analysis is usually conducted under a single probability measure. However, in practice, the available evidence often pins the measure down only to a set. Consequently, under a set of probability measures, VOI requires different formulations. First, we explicate a rule-specific VOI that fixes a decision rule for acting under imprecision (such as Gamma-maximin) and measures what the information is worth to a decision maker who uses that rule. Second, we derive a fixed-measure envelope that evaluates the classical VOI functional over all admissible precise measures. We formalize this distinction and explicate its consequences for the expected perfect, partial, and sample information. The expected value of perfect information is concave over the credal set. Hence, when the set is generated by finitely many measures, its lower envelope endpoint is obtained exactly from the generators, while its upper endpoint may be interior and is computed by a finite linear program. The Gamma-maximin value, in contrast, can exceed the entire envelope, so a rule-specific value is not recovered from the envelope's endpoints. A continuity bound limits how much the VOI can change as the measure varies, and we identify when the partial- and sample-information endpoints can still be obtained from the generators. Because the single-measure VOI must itself be estimated, the procedure we give combines standard estimators for it with a search over the credal set. By using a worked decision problem, we show how the two quantities separate conclusions that hold across every admissible measure from conclusions that depend on one unidentified choice of measure.

Online high-dimensional regression requires algorithms that can update sequentially while preserving structural sparsity. We propose Adaptive Iterative Hard Thresholding (AIHT), an online sparse-regression framework that alternates stochastic subgradient updates with adaptively scheduled hard-thresholding steps. The key idea is to separate support discovery from local refinement: early in the learning process, AIHT delays thresholding so that weak but informative coordinates have time to accumulate signal, while later it increases the projection frequency to stabilize the sparse estimator and exploit local curvature. We develop the theory for high-dimensional online quantile regression, a challenging setting in which the loss is nonsmooth and the data may exhibit heterogeneity or heavy-tailed noise. Under restricted curvature and gradient-leakage conditions, AIHT remains in an inflated sparse cone, exhibits a two-phase convergence behavior, and attains logarithmic regret for the sliding-window objective. Simulations for online quantile regression, together with threshold-scheduling ablations, support the proposed mechanism and illustrate its advantage over standard online sparse-learning baselines.

Considering pointwise and sup-norm estimation, we analyze the non-asymptotic behavior of local averaging estimators for Lipschitz regression functions. Building on a general deviation bound for estimators based on a VC family of localizing sets, we introduce the notion of shape-regular local maps, where averaging is performed over sets with an almost isotropic geometry. Our main message is a characterization: shape regularity is both necessary and sufficient to attain optimal rates, up to logarithmic factors. Necessity is established non-asymptotically through an explicit anisotropic example, sharpening a phenomenon previously understood only heuristically in asymptotic theory. We then draw two consequences. First, the simple $k$-nearest neighbor rule is shape-regular by construction and attains the optimal rate, even on unbounded supports. Second, and perhaps surprisingly, the popular random-split condition for trees – known to ensure consistency and vanishing cell diameters – does not guarantee optimal rates: for blind tree constructions, the cell aspect ratio diverges exponentially with depth, so that shape regularity fails with positive probability. This identifies the absence of a geometric correction mechanism, rather than a slowly shrinking diameter, as the obstruction to optimality. Motivated by this gap, we propose a tree construction that enforces shape regularity through a simple constraint on admissible splits, and prove a uniform deviation inequality showing that it restores the optimal rate for Lipschitz functions.