Papers

Precise recall control is critical in large-scale spatial conflation and entity-matching tasks, where missing even a few true matches can break downstream analytics, while excessive manual review inflates cost. Classical confidence-interval cuts such as Clopper-Pearson or Wilson provide lower bounds on recall, but they routinely overshoot the target by several percentage points and exhibit high run-to-run variance under skewed score distributions. We present an end-to-end framework that achieves exact recall with sub-percent variance over tens of millions of geometry pairs, while remaining TPU-friendly. Our pipeline starts with an equigrid bounding-box filter and compressed sparse row (CSR) candidate representation, reducing pair enumeration by two orders of magnitude. A deterministic xxHash bootstrap sample trains a lightweight neural ranker; its scores are propagated to all remaining pairs via a single forward pass and used to construct a reproducible, score-decile-stratified calibration set. Four complementary threshold estimators - Clopper-Pearson, Jeffreys, Wilson, and an exact quantile - are aggregated via inverse-variance weighting, then fused across nine independent subsamples. This ensemble reduces threshold variance compared to any single method. Evaluated on two real cadastral datasets (approximately 6.31M and 67.34M pairs), our approach consistently hits a recall target within a small error, decreases redundant verifications relative to other calibrations, and runs end-to-end on a single TPU v3 core.

Smart wearables enable continuous tracking of established biomarkers such as heart rate, heart rate variability, and blood oxygen saturation via photoplethysmography (PPG). Beyond these metrics, PPG waveforms contain richer physiological information, as recent deep learning (DL) studies demonstrate. However, DL models often rely on features with unclear physiological meaning, creating a tension between predictive power, clinical interpretability, and sensor design. We address this gap by introducing PPGen, a biophysical model that relates PPG signals to interpretable physiological and optical parameters. Building on PPGen, we propose hybrid amortized inference (HAI), enabling fast, robust, and scalable estimation of relevant physiological parameters from PPG signals while correcting for model misspecification. In extensive in-silico experiments, we show that HAI can accurately infer physiological parameters under diverse noise and sensor conditions. Our results illustrate a path toward PPG models that retain the fidelity needed for DL-based features while supporting clinical interpretation and informed hardware design.

A central challenge in Bayesian inference is efficiently approximating posterior distributions. Stein Variational Gradient Descent (SVGD) is a popular variational inference method which transports a set of particles to approximate a target distribution. The SVGD dynamics are governed by a reproducing kernel Hilbert space (RKHS) and are highly sensitive to the choice of the kernel function, which directly influences both convergence and approximation quality. The commonly used median heuristic offers a simple approach for setting kernel bandwidths but lacks flexibility and often performs poorly, particularly in high-dimensional settings. In this work, we propose an alternative strategy for adaptively choosing kernel parameters over an abstract family of kernels. Recent convergence analyses based on the kernelized Stein discrepancy (KSD) suggest that optimizing the kernel parameters by maximizing the KSD can improve performance. Building on this insight, we introduce Adaptive SVGD (Ad-SVGD), a method that alternates between updating the particles via SVGD and adaptively tuning kernel bandwidths through gradient ascent on the KSD. We provide a simplified theoretical analysis that extends existing results on minimizing the KSD for fixed kernels to our adaptive setting, showing convergence properties for the maximal KSD over our kernel class. Our empirical results further support this intuition: Ad-SVGD consistently outperforms standard heuristics in a variety of tasks.

Normalising-flow variational inference (VI) can approximate complex posteriors, yet single-flow models often behave inconsistently across qualitatively different distributions. We propose Adaptive Mixture Flow Variational Inference (AMF-VI), a heterogeneous mixture of complementary flows (MAF, RealNVP, RBIG) trained in two stages: (i) sequential expert training of individual flows, and (ii) adaptive global weight estimation via likelihood-driven updates, without per-sample gating or architectural changes. Evaluated on six canonical posterior families of banana, X-shape, two-moons, rings, a bimodal, and a five-mode mixture, AMF-VI achieves consistently lower negative log-likelihood than each single-flow baseline and delivers stable gains in transport metrics (Wasserstein-2) and maximum mean discrepancy (MDD), indicating improved robustness across shapes and modalities. The procedure is efficient and architecture-agnostic, incurring minimal overhead relative to standard flow training, and demonstrates that adaptive mixtures of diverse flows provide a reliable route to robust VI across diverse posterior families whilst preserving each expert's inductive bias.

Improving statistical forecasts of tropical cyclone (TC) intensity is limited by complex nonlinear interactions and difficulty in identifying relevant predictors. Conventional methods prioritize correlation or fit, often overlooking confounding variables and limiting generalizability to unseen TCs. To address this, we leverage a multidata causal discovery framework with a replicated dataset based on Statistical Hurricane Intensity Prediction Scheme (SHIPS) using ERA5 meteorological reanalysis. We conduct experiments to identify and select predictors causally linked to TC intensity changes. We then train multiple linear regression models to compare causal feature selection with correlation, random forest feature importance, and no feature selection, across five forecast lead times from 1 to 5 days (24 to 120 hours). Causal feature selection consistently outperforms on unseen test cases, especially for lead times shorter than 3 days. Top causal features include vertical shear, mid-tropospheric potential vorticity and surface moisture conditions, which are physically significant yet often underutilized in TC intensity predictions. We build an extended predictor set (SHIPS+) by adding selected features to the standard SHIPS predictors. SHIPS+ yields increased short-term predictive skill at lead times of 24, 48, and 72 hours. Adding nonlinearity using a multilayer perceptron further extends skill to longer lead times, despite our framework being purely regional and not requiring global forecast data. Operational SHIPS tests confirm that three of the six added causally discovered predictors improve forecast skill, with the largest gains at longer lead times. Our results demonstrate that causal discovery improves TC intensity prediction and pave the way toward more empirical forecasts.

We consider adversarially robust classification in a multiclass setting under arbitrary loss functions and derive dual and barycentric reformulations of the corresponding learner-agnostic robust risk minimization problem. We provide explicit characterizations for important cases such as the cross-entropy loss, loss functions with a power form, and the quadratic loss, extending in this way available results for the 0-1 loss. These reformulations enable efficient computation of sharp lower bounds for adversarial risks and facilitate the design of robust classifiers beyond the 0-1 loss setting. Our paper uncovers interesting connections between adversarial robustness, $\alpha$-fair packing problems, and generalized barycenter problems for arbitrary positive measures where Kullback-Leibler and Tsallis entropies are used as penalties. Our theoretical results are accompanied with illustrative numerical experiments where we obtain tighter lower bounds for adversarial risks with the cross-entropy loss function.

We propose a continuous-time formulation of a noisy persistent contrastive divergence (PCD)-like method for maximum likelihood estimation (MLE) of unnormalised densities. Our approach couples parameter updates and sampling of the parametrised density in a multiscale system of stochastic differential equations (SDEs). From this formulation, we derive non-asymptotic bounds for weak test-function errors between the resulting numerical schemes and the MLE point target. The error is decomposed into numerical discretisation, slow-fast averaging, and finite-temperature concentration terms. We also introduce an efficient implementation based on explicit stabilized integrators and establish corresponding long-time error estimates. This leads to a novel method for training energy-based models (EBMs) with quantitative error guarantees.

Given a known function $f : [0, 1] \to (0, 1)$ and a random but almost surely finite number of independent, Ber$(x)$-distributed random variables with unknown $x \in [0, 1]$, we prove the existence of an unbiased, $[0, 1]$-valued estimator of the probability $f(x) \in (0, 1)$. Our estimator is based on so-called debiasing, or randomly truncating a telescopic series of consistent estimators. Debiased estimators of a probability are not typically constrained to $[0, 1]$, or even bounded, even when all consistent estimators used as inputs are. We show that constructing the series of consistent estimators from the coefficients of a particular Bernoulli factory yields provable boundedness provided $f \in C^{\rho}[0, 1]$ for $\rho > 3$. Our result can be thought of as a novel Bernoulli factory with the appealing property that the required number of Ber$(x)$-distributed random variates is independent of their outcomes.

Diagonal linear networks (DLNs) are a tractable model that captures several nontrivial behaviors in neural network training, such as initialization-dependent solutions and incremental learning. These phenomena are typically studied in isolation, leaving the overall dynamics insufficiently understood. In this work, we present a unified analysis of various phenomena in the gradient flow dynamics of DLNs. Using Dynamical Mean-Field Theory (DMFT), we derive a low-dimensional effective process that captures the asymptotic gradient flow dynamics in high dimensions. Analyzing this effective process yields new insights into DLN dynamics, including loss convergence rates and their trade-off with generalization, and systematically reproduces many of the previously observed phenomena. These findings deepen our understanding of DLNs and demonstrate the effectiveness of the DMFT approach in analyzing high-dimensional learning dynamics of neural networks.

We advocate for a new statistical principle that combines the most desirable aspects of both parameter inference and density estimation. This leads us to the predictively oriented (PrO) posterior, which expresses uncertainty as a consequence of predictive ability. We show that these posteriors converge to the predictively optimal model average and predictively dominate both classical and generalised Bayes posterior predictive distributions. Further, PrO posteriors adapt to the level of model misspecification: while they concentrate around the true model in the same way as classical and generalised Bayesian strategies if the model can recover the data-generating distribution, they do not concentrate around a single model in the presence of non-trivial forms of model misspecification. Instead, they stabilise towards a non-degenerate predictively optimal posterior distribution that represents a form of irreducible uncertainty due to model misspecification. We put forward a sampling algorithm for PrO posteriors based on mean field Langevin dynamics, and verify the practical significance of our theoretical developments on a number of numerical examples.

Sequential Monte Carlo algorithms, or particle filters, are widely used for approximating intractable integrals, particularly those arising in Bayesian inference and state-space models. We introduce a new variance reduction technique, the knot operator, which improves the efficiency of particle filters by incorporating potential function information into part, or all, of a transition kernel. The knot operator induces a partial ordering of Feynman-Kac models that implies an order on the asymptotic variance of particle filters, offering a new approach to algorithm design. We discuss connections to existing strategies for designing efficient particle filters, including model marginalisation. Our theory generalises such techniques and provides quantitative asymptotic variance ordering results. We revisit the fully-adapted (auxiliary) particle filter using our theory of knots to show how a small modification guarantees an asymptotic variance ordering for all relevant test functions.

This paper examines replication portfolio construction in incomplete markets - a key problem in financial engineering with applications in pricing, hedging, balance sheet management, and energy storage planning. We model this as a two-player game between an investor and the market, where the investor makes strategic bets on future states while the market reveals outcomes. Inspired by the success of Monte Carlo Tree Search in stochastic games, we introduce an AlphaZero-based system and compare its performance to deep hedging - a widely used industry method based on gradient descent. Through theoretical analysis and experiments, we show that deep hedging struggles in environments where the optimal action-value function is not subject to convexity constraints - such as those involving non-convex transaction costs, capital constraints, or regulatory limitations - converging to local optima. We construct specific market environments to highlight these limitations and demonstrate that AlphaZero consistently finds near-optimal replication strategies. On the theoretical side, we establish a connection between deep hedging and convex optimization, suggesting that its effectiveness is contingent on convexity assumptions. Our experiments further suggest that AlphaZero is more sample-efficient - an important advantage in data-scarce, overfitting-prone derivative markets.

To address the common problem of high dimensionality in tensor regressions, we introduce a generalized tensor random projection method that embeds high-dimensional tensor-valued covariates into low-dimensional subspaces with minimal loss of information about the responses. The method is flexible, allowing for tensor-wise, mode-wise, or combined random projections as special cases. A Bayesian inference framework is provided featuring the use of a hierarchical prior distribution and a low-rank representation of the parameter. Strong theoretical support is provided for the concentration properties of the random projection and posterior consistency of the Bayesian inference. An efficient Gibbs sampler is developed to perform inference on the compressed data. To mitigate the sensitivity introduced by random projections, Bayesian model averaging is employed, with normalising constants estimated using reverse logistic regression. An extensive simulation study is conducted to examine the effects of different tuning parameters. Simulations indicate, and the real data application confirms, that compressed Bayesian tensor regression can achieve better out-of-sample prediction while significantly reducing computational cost compared to standard Bayesian tensor regression.

Designing categorical kernels is a major challenge for Gaussian process regression with continuous and categorical inputs. Despite previous studies, it is difficult to identify a preferred method, either because the evaluation metrics, the optimization procedure, or the datasets change depending on the study. In particular, reproducible code is rarely available. The aim of this paper is to provide a reproducible comparative study of all existing categorical kernels on many of the test cases investigated so far. We also propose new evaluation metrics inspired by the optimization community, which provide quantitative rankings of the methods across several tasks. From our results on datasets which exhibit a group structure on the levels of categorical inputs, it appears that nested kernels methods clearly outperform all competitors. When the group structure is unknown or when there is no prior knowledge of such a structure, we propose a new clustering-based strategy using target encodings of categorical variables. We show that on a large panel of datasets, which do not necessarily have a known group structure, this estimation strategy still outperforms other approaches while maintaining low computational cost.

The increase in the collection of biological data allows for the individual and longitudinal monitoring of hematological or urine biomarkers. However, identifying abnormal behavior in these biological sequences is not trivial. Moreover, the complexity of the biological data (correlation between biomarkers, seasonal effects, etc.) is also an issue. Z-score methods can help assess the abnormality in these longitudinal sequences while capturing some features of the biological complexity. This work details a statistical framework for handling biological sequences using three custom Z-score methods in the intra-individual variability scope. These methods can detect abnormal samples in the longitudinal sequences with respect to the seasonality, chronological time or correlation between biomarkers. One of these methods is an extension of one custom Z-score method to the Gaussian linear model, which allows for including additional variables in the model design. We illustrate the use of the framework on the longitudinal data of 3,936 professional soccer players (5 biomarkers) and 1,683 amateur or professional cyclists (10 biomarkers). The results show that a particular Z-score method, designed to detect a change in a series of consecutive observations, measured a high proportion of abnormal values (more than three times the false positive rate) in the ferritin and IGF1 biomarkers for both data sets. The proposed framework and methods could be applied in other contexts, such as the clinical patient follow-up in monitoring abnormal values of biological markers. The methods are flexible enough to include more complicated biological features, which can be directly incorporated into the model design.

Death events commonly arise from complex interactions among interrelated causes, formally classified in reporting practices as underlying and contributing. Leveraging information from death certificates, these interactions can be naturally represented through a sequence of directed networks encoding co-occurrence strengths between pairs of underlying and contributing causes across ages. Although this perspective opens the avenues to learn informative age-specific block interactions among endogenous groups of underlying and contributing causes displaying similar co-occurrence patterns, there has been limited research along this direction in mortality modeling. This is mainly due to the lack of suitable stochastic block models for age-indexed sequences of directed networks. We cover this gap through a novel Bayesian formulation which crucially learns two separate group structures for underlying and contributing causes, while allowing both structures to change smoothly across ages via dependent random partition priors. As illustrated in simulations, this formulation outperforms state-of-the-art solutions that could be adapted to our motivating application. Moreover, when applied to USA mortality data, it unveils structures in the composition, evolution, and modular interactions among causes-of-death groups that were hidden to previous studies. Such findings could have relevant policy implications and contribute to an improved understanding of the recent "death of despair" phenomena in USA.

Understanding how individuals perceive their living environment is a complex task, as it reflects both personal and contextual determinants. In this paper, we address this task by analyzing the environmental module of the Italian nationwide health surveillance system PASSI (Progressi delle Aziende Sanitarie per la Salute in Italia), integrating it with contextual information at the municipal level, including socio-economic indicators, pollution exposure, and other geographical characteristics. Methodologically, we adopt a penalized semi-parallel cumulative ordinal regression model to analyze how subjective perceptions are shaped by both personal and territorial determinants. The approach balances flexibility and interpretability by allowing both parallel and non-parallel effects while regularizing estimates to address multicollinearity and separation issues. We use the model as an analytical tool to uncover the determinants of positivity and neutrality in environmental perceptions, defined as factors that contribute the most to improving perception or increasing the sense of neutrality. The results are diverse. First, results reveal significant heterogeneity across Italian territories, indicating that local characteristics strongly shape environmental perception. Second, various individual factors interact with contextual influences to shape perceptions. Third, hazardous environmental factors, such as higher PM2.5 levels, appear to be associated with poorer environmental perception, suggesting a tendency among respondents to recognize specific environmental issues. Overall, the approach demonstrates strong potential for application and provides useful insights for environmental policy planning.

The Eilers-Whittaker method for data smoothing effectiveness depends on the choice of the regularisation parameter, and automatic selection is a necessity for large datasets. Common methods, such as leave-one-out cross-validation, can perform poorly when serially correlated noise is present. We propose a novel procedure for selecting the control parameter based on the spectral entropy of the residuals. We define an S-curve from the Euclidean distance between points in a plot of the spectral entropy of the residuals versus that of the smoothed signal. The regularisation parameter corresponding to the absolute maximum of this S-curve is chosen as the optimal parameter. Using simulated data, we benchmarked our method against cross-validation and the V-curve. Validation was also performed on diverse experimental data. This robust and straightforward procedure can be a valuable addition to the available selection methods for the Eilers smoother.

Spatial data are central to applications such as environmental monitoring and urban planning, but are often distributed across devices where privacy and communication constraints limit direct sharing. Federated modeling offers a practical solution that preserves data privacy while enabling global modeling across distributed data sources. For instance, environmental sensor networks are privacy- and bandwidth-constrained, motivating federated spatial modeling that shares only privacy-preserving summaries to produce timely, high-resolution pollution maps without centralizing raw data. However, existing federated modeling approaches either ignore spatial dependence or rely on synchronous updates that suffer from stragglers in heterogeneous environments. This work proposes an asynchronous federated modeling framework for spatial data based on low-rank Gaussian process approximations. The method employs block-wise optimization and introduces strategies for gradient correction, adaptive aggregation, and stabilized updates. We establish linear convergence with explicit dependence on staleness, a result of standalone theoretical significance. Moreover, numerical experiments demonstrate that the asynchronous algorithm achieves synchronous performance under balanced resource allocation and significantly outperforms it in heterogeneous settings, showcasing superior robustness and scalability.

This paper primarily considers the robust estimation problem under Wasserstein distance constraints on the parameter and noise distributions in the linear measurement model with additive noise, which can be formulated as an infinite-dimensional nonconvex minimax problem. We prove that the existence of a saddle point for this problem is equivalent to that for a finite-dimensional minimax problem, and give a counterexample demonstrating that the saddle point may not exist. Motivated by this observation, we present a verifiable necessary and sufficient condition whose parameters can be derived from a convex problem and its dual. Additionally, we also introduce a simplified sufficient condition, which intuitively indicates that when the Wasserstein radii are small enough, the saddle point always exists. In the absence of the saddle point, we solve an finite-dimensional nonconvex minimax problem, obtained by restricting the estimator to be linear. Its optimal value establishes an upper bound on the robust estimation problem, while its optimal solution yields a robust linear estimator. Numerical experiments are also provided to validate our theoretical results.

Response-adaptive randomization (RAR) methods can be used to adapt randomization probabilities based on accumulating data, aiming to increase the probability of allocating patients to effective treatments. A popular RAR method is Thompson sampling, which randomizes patients proportionally to the Bayesian posterior probability that each treatment is the most effective. However, its high variability can also increase the risk of assigning patients to inferior treatments and lead to inferential problems such as confidence interval undercoverage. We propose a principled method based on Bayesian hypothesis testing to address these issues: We introduce a null hypothesis postulating equal effectiveness of treatments. Bayesian model averaging then induces shrinkage toward equal randomization probabilities, with the degree of shrinkage controlled by the prior probability of the null hypothesis. Equal randomization and Thompson sampling arise as special cases when the prior probability is one or zero, respectively. A simulation study demonstrates that the method can mitigate issues with Thompson sampling and has comparable statistical properties to Thompson sampling with common ad hoc modifications such as power transformation and probability capping. Under the null hypothesis and a normal model, the randomization probabilities are shown to converge asymptotically to equal randomization, unlike those of Thompson sampling. We implement the method in the free and open-source R package brar, enabling experimenters to easily perform null hypothesis Bayesian RAR and support more effective randomization of patients.

Particulate matter data now include various particle sizes, which often manifest as a collection of curves observed sequentially over time. When considering 51 distinct particle sizes, these curves form a high-dimensional functional time series observed over equally spaced and densely sampled grids. While high dimensionality poses statistical challenges due to the curse of dimensionality, it also offers a rich source of information that enables detailed analysis of temporal variation across short time intervals for all particle sizes. To model this complexity, we propose a multilevel functional time series framework incorporating a functional factor model to facilitate one-day-ahead forecasting. To quantify forecast uncertainty, we develop a calibration approach and a split conformal prediction approach to construct prediction intervals. Both approaches are designed to minimise the absolute difference between empirical and nominal coverage probabilities using a validation dataset. Furthermore, to improve forecast accuracy as new intraday data become available, we implement dynamic updating techniques for point and interval forecasts. The proposed methods are validated through an empirical application to hourly measurements of particulate matter in 51 size categories in London.

Image denoising is a fundamental problem in computer vision and medical imaging. However, real-world images are often degraded by structured noise with strong anisotropic correlations that existing methods struggle to remove. Most data-driven approaches rely on large datasets with high-quality labels and still suffer from limited generalizability, whereas existing zero-shot methods avoid this limitation but remain effective only for independent and identically distributed (i.i.d.) noise. To address this gap, we propose Median2Median (M2M), a zero-shot denoising framework designed for structured noise. M2M introduces a novel sampling strategy that generates pseudo-independent sub-image pairs from a single noisy input. This strategy leverages directional interpolation and generalized median filtering to adaptively exclude values distorted by structured artifacts. To further enlarge the effective sampling space and eliminate systematic bias, a randomized assignment strategy is employed, ensuring that the sampled sub-image pairs are suitable for Noise2Noise training. In our realistic simulation studies, M2M performs on par with state-of-the-art zero-shot methods under i.i.d. noise, while consistently outperforming them under correlated noise. These findings establish M2M as an efficient, data-free solution for structured noise suppression and mark the first step toward effective zero-shot denoising beyond the strict i.i.d. assumption.

A popular task in generalization is to learn about a new, target population based on data from an existing, source population. This task relies on conditional exchangeability, which asserts that differences between the source and target populations are fully captured by observable characteristics of the two populations. Unfortunately, this assumption is often untenable in practice due to unobservable differences between the source and target populations. Worse, the assumption cannot be verified with data, warranting the need for robust data collection processes and study designs that are inherently less sensitive to violation of the assumption. In this paper, we propose SLOPE (Sensitivity of LOcal Perturbations from Exchangeability), a simple, intuitive, and novel measure that quantifies the sensitivity to local violation of conditional exchangeability. SLOPE combines ideas from sensitivity analysis in causal inference and derivative-based measure of robustness from Hampel (1974). Among other properties, SLOPE can help investigators to choose (a) a robust source or target population or (b) a robust estimand. Also, we show an analytic relationship between SLOPE and influence functions, which investigators can use to derive SLOPE given an influence function. We conclude with a re-analysis of a multi-national randomized experiment and illustrate the role of SLOPE in informing robust study designs for generalization.

This paper introduces an Artificial Intelligence (AI) foundation model for time series in engineering applications, where causal operations are required for real-time monitoring and control. Since engineering time series are governed by physical, rather than linguistic, laws, large-language-model-based AI foundation models may be ineffective or inefficient. Building on the classical innovations representation theory of Wiener, Kallianpur, and Rosenblatt, we propose Time Series GPT (TS-GPT) – an innovations-representation-based Generative Pre-trained Transformer for engineering monitoring and control. As an example of foundation model adaptation, we consider Probabilistic Generative Forecasting, which produces future time series samples from conditional probability distributions given past realizations. We demonstrate the effectiveness of TS-GPT in forecasting real-time locational marginal prices using historical data from U.S. independent system operators.