Papers

We propose a physically explicit sensing application of a time-reversed Young (TRY) interferometer: simultaneous monitoring of beam tilt and focus drift with a fixed detector. The task is relevant to compact optical relays, free-space links, fiber-coupling stages, and micro-optical alignment modules, where continuous tracking of pointing and focus is needed but downstream wavefront cameras or multiport analyzers are undesirable. Using a finite-width double-slit Fresnel model, we derive the exact local TRY response functions for tilt-like and defocus-like phase perturbations and compute the corresponding optimal upstream source codes numerically. The physical optimal codes are fringe-locked and differ qualitatively from the simple odd/even modes suggested by Gaussian toy models. Two source-coded scalar channels recover essentially all local Fisher information in the full source-resolved TRY record for the physical model considered here. Compared with downstream direct intensity sensing, TRY provides first-order access to the mixed tilt–defocus task with fixed detection; compared with ideal downstream matched-mode sorting, its advantage is architectural rather than fundamental.

In a recent paper \cite{Redei-Jing2026} the notion of conditional $p$-inaccessibility of a decision based on utility maximization was defined and examples of conditionally $p$-inaccessible decisions were given. The conditional inaccessibility of a decision based on maximizing utility calculated by a probability measure $p^*$ expresses that the decision cannot be obtained if the expectation values of the utility functions are calculated using the (Jeffrey) conditional probability measure obtained by conditioning $p$ on partial evidence about the probability $p^*$ that determines the decision. The paper \cite{Redei-Jing2026} conjectured that conditionally $p$-inaccessible decisions exist in some probability spaces having arbitrary large finite number of elementary events. In this paper we prove that for any $p$ in any finite probability space there exist an uncountable number of probability measures $p^*$ for each of which there exist an uncountable number of pairs of utility functions that represent conditionally $p$-inaccessible decisions. If $p^*$ is an objective probability determining objectively good decisions and $p$ is the subjective probability determining a rational decision of a decision making Agent, the result says that there is an enormous number of decision situations in which the Agent's subjective probability prohibits the Agent's informed rational decision to be objectively good.

Let $\mathcal{C}_a$ denote the class of associative copulas, and let $\overline{\mathcal{C}}_a$ be the closure, in the uniform metric $d_\infty$, of the convex hull of $\mathcal{C}_a$ . It is known that $\mathcal{C}_a \subseteq \mathcal{C}_{SC}$, the class of Schur-concave commutative copulas. We prove the reverse inclusion, establishing $\overline{\mathcal{C}}_a = \mathcal{C}_{SC}$.

Rare disease trials face unique statistical challenges due to limited patient populations and heterogeneous clinical manifestations among patients. Multiple endpoints are often necessary to comprehensively capture treatment benefits. A global test is an approach for evaluating whether a treatment has any beneficial effect across multiple endpoints. We propose a new global test based on a weighted composite endpoint. The proposed global test employs shrinkage-based cross-validated targeted maximum likelihood estimation (CV-TMLE) to learn data-adaptive weights that maximize power while maintaining Type I error control. Shrinkage can be tailored to incorporate existing domain knowledge, such as anticipated relative effect sizes. In simulation studies designed to reflect real rare disease trial settings, the proposed procedure demonstrated improved power over standard multiplicity adjustments and classical global tests (such as the O'Brien test), while maintaining nominal Type I error, when effects are heterogeneous across endpoints. The proposed method simultaneously learns an optimal weighted composite outcome and provides an unbiased and efficient targeted maximum likelihood estimator (TMLE) for the average treatment effect (ATE) on that weighted outcome, with valid inference taking into account that the ATE is data dependent.

Probabilistic values, including Shapley values and semivalues, provide a model-agnostic framework to attribute the behavior of a black-box model to data points or features, with a wide range of applications including explainable artificial intelligence and data valuation. However, their exact computation requires utility evaluations over exponentially many coalitions, making Monte Carlo approximation essential in modern machine learning applications. Existing estimators are often developed through different representation strategies, including weighted averages, self-normalized weighting, regression adjustment, and weighted least squares. Our key observation is that these seemingly distinct constructions share a common first-order expansion, in which the leading term is determined by the sampling law and a working surrogate function. This first-order representation yields an explicit expression for the leading mean squared error (MSE), which characterizes how the sampling law and the surrogate jointly determine statistical efficiency. Guided by this criterion, we propose an Efficiency-Aware Surrogate-adjusted Estimator (EASE) that directly chooses the sampling law and surrogate to minimize the first-order MSE. We demonstrate that EASE consistently outperforms existing estimators for various probabilistic values.

This chapter introduces the Bayesian reflex – an analogy with the autonomic nervous system – as a unifying framework for online learning in AI. Bayesian online algorithms automatically maintain equilibrium in dynamic environments via three mechanisms: belief maintenance through probabilistic representations, sequential updating via Bayes' theorem, and uncertainty-driven action balancing exploration and exploitation. We survey online Bayesian methods, highlighting two computational principles: the look-up table principle for sequential inference in function space, and the ellipsoidal decomposition framework for nearly exact i.i.d. sampling from arbitrary posteriors. These principles are generalized across dynamic emulation, nonparametric state-space models, circular time series, inverse regression for climate model evaluation, and deep architectures via Recursive Gaussian Processes. Decision-making is explored via Thompson sampling and restless bandits. We extend the framework to assess infinite series convergence (applied to climate dynamics and the Riemann Hypothesis), model prime number distributions leading to the discovery of 184 strong Mersenne prime candidates, detect stationarity, and characterize point processes. The Bayesian reflex provides a foundational infrastructure for adaptive AI that continuously learns in a complex world.

Day-to-day traffic dynamics are widely used to model flow evolution due to travelers' learning and adjustment behavior, yet empirical analysis of these models often relies on descriptive calibration with limited inferential content. This paper develops a statistical inference framework for day-to-day route choice dynamics based on a stochastic individual-level adjustment model. The framework enables uncertainty quantification and formal inference for behavioral parameters from trajectory data. We establish identifiability and consistency under mild conditions, and extend the framework to accommodate demand variation, user heterogeneity through a hierarchical structure, and anonymized observability caused by privacy constraints on trajectory data. Simulation studies demonstrate good finite-sample performance, calibrated uncertainty, and robustness to model misspecification. Empirical analyses of controlled laboratory experiments and real-world trajectory data from Ann Arbor, Michigan, show that the framework can generate novel behavioral insights across settings: it reveals the inadequacy of a purely inter-day learning model once en-route information is introduced, recovers systematic behavioral differences across participant types, and uncovers meaningful day-to-day learning together with substantial demand variation in real-world commuting behavior.

We consider the infinite-width limit of a fully connected deep neural network with general weights, and we prove quantitative general bounds on the $2$-Wasserstein distance between the network and its infinite-width Gaussian limit, under appropriate regularity assumptions on the activation function. Our main tool is a Lindeberg principle for Deep Neural Networks, which we use to successively replace the weights on each layer by Gaussian random variables.

The COVID-19 pandemic has been characterised by multiple waves of transmission driven by interventions and emerging variants, challenging epidemic models that assume gradually evolving transmission dynamics. We propose a class of state-space models in which the transmission rate evolves through persistent regimes of random duration, governed by a semi-Markov process. This formulation yields an interpretable representation of sustained transmission phases and retains a parsimonious parameterisation. Particle-based Bayesian methods are well established for standard state-space models, but their use in semi-Markov settings has received comparatively limited attention. In epidemic applications, inference is further complicated by differential equation-driven latent dynamics and observation models defined through functionals of the latent process. We develop an inferential framework that accommodates these features, combining particle-based state updates with gradient-based parameter updates and enabling batch and sequential inference via particle and sequential Monte Carlo. We apply the proposed methodology to COVID-19 data from the United Kingdom and show that combining reported cases and deaths leads to more precise and stable inference compared to using deaths alone. These results illustrate the practical value of semi-Markov transmission models for epidemic analysis under complex observation schemes.

We propose a robust gradient estimator based on per-sample gradient clipping and analyze its properties both theoretically and empirically. We show that the resulting method, per-sample clipped SGD (PS-Clip-SGD), achieves optimal in-expectation convergence rates for non-convex optimization problems under heavy-tailed gradient noise. Moreover, we establish high-probability convergence guarantees that match the in-expectation rates up to polylogarithmic factors in the failure probability. We complement our theoretical results with multiple numerical experiments. In particular, we demonstrate that PS-Clip-SGD outperforms both vanilla SGD with momentum and standard gradient clipping when training AlexNet on the CIFAR-100 dataset, even after accounting for the additional computational time caused by per-sample clipping. We also empirically show that, in the presence of gradient accumulation, applying clipping at the mini-batch level can improve training performance while incurring virtually no additional computational cost. This finding is particularly interesting, as it contradicts the common practice of applying clipping only after all accumulation steps have been completed.

Across many scientific disciplines, multiple observations are collected from the same experimental units, and in modern datasets these observations often arise as non-Euclidean random objects. In such settings, the incorporation of random effects is a critical modeling step for efficient estimation and personalized prediction. Although mixed-effects models are well established for scalar outcomes and, more recently, for functional data in Hilbert spaces, general random-effects frameworks for objects in metric spaces remain underdeveloped. In this paper, we propose a nonlinear Fréchet-based algorithm for random-effects modeling of arbitrary random objects defined on a metric space. Using M-estimation theory, we establish conditions under which the proposed metric-space prediction target is consistently estimated under a working random-effects formulation. We then evaluate the empirical performance of the proposed method using both synthetic data and digital health datasets that require practical tools for analyzing random objects in metric spaces, such as multivariate probability distributions and random graphs. We show that, although our method is developed beyond Hilbert spaces, it can outperform existing Hilbert space-based methods.

The proliferation of large-scale and structurally complex data has spurred the integration of machine learning methods into statistical modeling. Recurrent neural networks (RNNs), a foundational class of models for time-dependent data, can be viewed as nonlinear extensions of classical autoregressive moving average models. Despite their flexibility and empirical success in machine learning, RNNs often suffer from limited interpretability and slow training, which hinders their use in statistics. This paper proposes the Parallelized RNN (ParaRNN), a novel model composed of multiple small recurrent units. ParaRNN admits an additive representation that decouples recurrent dynamics into interpretable components, whose behavior can be characterized through recurrence features. This interpretability enables its applications in nonparametric regression for time-dependent data, while the design also allows efficient parallelization. The approximation capacity and non-asymptotic prediction error bounds in a nonparametric regression setting are established for ParaRNN. Empirical results on three sequential modeling tasks further demonstrate that ParaRNN achieves performance comparable to vanilla RNNs while offering improved interpretability and efficiency.

This paper introduces an extension of generalised filtering for online applications. Generalised filtering refers to data assimilation schemes that jointly infer latent states, learn unknown model parameters, and estimate uncertainty in an integrated framework – e.g., estimate state and observation noise – at the same time (i.e., triple estimation). This framework appears across disciplines under different names, including variational Kalman-Bucy filtering in engineering, generalised predictive coding in neuroscience, and Dynamic Expectation Maximisation (DEM) in time-series analysis. Here, we specialise DEM for “online” data assimilation, through a separation of temporal scales. We describe the variational principles and procedures that allow one to assimilate data in a way that allows for a slow updating of parameters and precisions, which contextualise fast Bayesian belief updating about the dynamic hidden states. Using numerical studies, we demonstrate the validity of online DEM (ODEM) using a non-linear – and potentially chaotic – generative model, to show that the ODEM scheme can track the latent states of the generative process, even when its functional form differs fundamentally from the dynamics of the generative model. Framed from a neuro-mimetic predictive coding perspective, ODEM offers a biologically inspired solution to online inference, learning, and uncertainty estimation in dynamic environments.

We describe the R package EstemPMM, which implements the Polynomial Maximization Method (PMM) for parameter estimation under non-Gaussian errors. PMM exploits higher-order cumulants of the error distribution – specifically the third standardized moment gamma_3 and fourth standardized moment gamma_4 – to construct estimators that outperform ordinary least squares (OLS) whenever the errors are asymmetric or leptokurtic. The package provides a unified interface for linear regression (lm_pmm2, lm_pmm3), autoregressive and moving-average time-series models (ar_pmm2, ma_pmm2, arma_pmm2, arima_pmm2, and seasonal variants), a data-driven dispatch function (pmm_dispatch) that automatically selects OLS, PMM2, or PMM3 based on the sample skewness and excess kurtosis, and Monte Carlo comparison utilities. The implementation uses R's S4 class system and follows standard generic interfaces (coef, fitted, residuals, predict, summary, AIC, logLik, vcov, confint). Asymptotic efficiency is characterised by Kunchenko-style coefficients g_2, g_3 in [0,1], defined as the ratios of the asymptotic variance of the PMM2 and PMM3 estimators to that of OLS. Monte Carlo experiments confirm the theoretical values and a WTI crude-oil case study illustrates the dispatcher and parameter-precision benefits of PMM2 on real heavy-tailed data. EstemPMM version 0.3.2 is available from CRAN at https://CRAN.R-project.org/package=EstemPMM under the GPL-3 licence.

The Hawkes process is used to model point process data where events occur in clusters and bursts. In a standard multivariate Hawkes process, every event that occurs in a dimension has an equal impact on the process intensity. However, this assumption is unrealistic in applications such as the modelling of message cascades where the effect of an event depends on whether it was the initiator or a member of a particular cluster. To alleviate this, we introduce a new Hawkes process model, the Ancestor Hawkes process, which allows the impact of each event to vary based on its origin. The relevance of the Ancestor Hawkes process is showcased on real data from a 9-person group chat, where our proposed approach reveals individual response preferences. Crucially, this is achieved in a privacy-conscious manner, as only the sender and the time at which a message was sent – but not its content – are utilised. These nuances of messaging cascades are missed by the standard Hawkes process, but are relevant for studying latent interaction structure and for personalised notification management.

Inference of brain functional connectivity networks from resting-state fMRI data is a key focus in neuroimaging. This paper introduces new Bayesian approaches for inferring a functional connectivity graph from multivariate resting-state fMRI time series of a single subject. Our methods rely on novel Bayesian priors on correlation matrices and a dedicated prior elicitation framework, which translates prior beliefs about the expected level and variability of correlations into interpretable hyperparameter choices, enabling the construction of expert-informed priors. When combined with a Gaussian likelihood, these priors also exhibit computational advantages. Compared to most existing methods for this problem that estimate constant weights, our model provides distributional weights defined by the posterior distributions for the connectivity graph, yielding more robust point estimates through the regularizing effect of expert-informed priors, evaluating uncertainty, and enabling a range of post-inference analyses. In particular, we derive a procedure for identifying significant connectivities based on posterior distributions of weights and credible sets. To the best of our knowledge, only one existing Bayesian functional connectivity model is applicable to single-subject resting-state fMRI data, making our approach a valuable addition to the field and demonstrating superior performance in our experiments.

Gaussian Process (GP) models provide a flexible framework for prediction and uncertainty quantification. For most covariance functions, however, exact GP prediction with $n$ points scales as $\mathcal{O}(n^3)$, making it prohibitively expensive for large datasets or large numbers of prediction points. While nearest neighbor-based prediction can work well in certain settings, non-pathological circumstances (like measurement noise, for example) can severely restrict its efficiency. This work presents a complementary approach where one conditions on carefully designed linear combinations of data, which is particularly effective in the setting of jointly predicting many values in large connected regions of the data domain. For kernel functions that are smooth away from the origin and simple prediction domains, this method can be exponentially convergent in the number of linear combinations $r$ used for conditioning. The procedure costs $\mathcal{O}(T r^2)$ work, where $T$ is the cost of solving a linear system with the data covariance matrix, and so in many cases can be computed in linear or near-linear cost by exploiting rank structure in well-behaved covariance matrices. At the cost of $\mathcal{O}(nr^2)$ additional precomputation work, this approach can also provide predictions at arbitrary points of a designated region in $\mathcal{O}(1)$ online work, making it particularly attractive for problems where prediction points are not known in advance. After establishing favorable theoretical properties, we provide several example applications to problems in prediction and matrix approximation.

Background and Objective: Uncertainty in non-linear mixed effect models is often assessed using the Fisher information matrix to derive the standard errors of estimation. The bootstrap is an alternative to the asymptotic method, with different approaches to handle the different levels of individual and population variabilities. The simplest method is the Case bootstrap where the entire vector of individuals is resampled, but this approach does not take into account the hierarchical nature of non-linear mixed effect models (NLMEM). Methods: We propose here a non-parametric bootstrap, cNP, to preserve the structure of the original data. We resample interindividual random effects from the conditional distribution of the individual parameters, obtained as a by-product of the SAEM algorithm, and residuals from their distribution. cNP was implemented in the saemix package for R along with the case, parametric (Par), and non-parametric (NP) residual bootstraps. Coverage rates were compared in a simulation study using sigmoid Emax models, with rich, sparse and unbalanced designs, and 3 levels of residual variability. Results: The asymptotic method tended to produce lower than theoretical coverages for the variance terms. Bootstraps provided more adequate coverage, but none of the approaches maintained coverage when the residual error increased. Overall, the new cNP and the Case provided better coverage than the classical NP. Conclusion: The new conditional non-parametric bootstrap can be used when it is important to preserve the structure of the original dataset, such as the number of observations or the repartition of covariates as it does not require stratification.

Background: Music-based interventions are increasingly used in neonatal intensive care units (NICUs), but the literature remains heterogeneous in intervention type, provider role, and research focus. This study examined research trends in NICU music-based intervention studies using text mining. Methods: We analyzed 83 abstracts from peer-reviewed studies published between 1998 and 2025. Methods included preprocessing, RAKE-based keyphrase extraction, keyword frequency analysis, temporal trend analysis, intervention-type comparison, and latent Dirichlet allocation topic modeling. The optimal number of topics was determined using the CaoJuan2009, Arun2010, and Deveaud2014 metrics. Results: Study volume increased steadily over time, with nearly half (38/83) published from 2020 onward. Early studies focused on passive music listening and short-term physiological outcomes, whereas recent studies increasingly examined singing, live music, and parent-involved interventions. Keyword analysis showed a shift from physiological stability and behavioral responses toward neurodevelopmental outcomes, parental emotional well-being, and parent-infant interaction. Music medicine studies emphasized passive auditory stimulation and immediate physiological outcomes, whereas music therapy studies addressed broader developmental, relational, and psychosocial topics. Topic modeling identified four major themes, with parent-involved physiological regulation and stress reduction the most frequent dominant topic. Conclusions: NICU music-based intervention research is becoming more interdisciplinary. The field has expanded from immediate physiological stabilization to broader developmental, relational, and psychosocial goals. Future work should clarify the distinction between music therapy and music medicine and promote interdisciplinary collaboration in NICU care.

Offline Reinforcement Learning from Human Feedback (RLHF) pipelines such as Direct Preference Optimization (DPO) train on a pre-collected preference dataset, which makes them vulnerable to preference poisoning attack. We study label flip attacks against log-linear DPO. We first illustrate that flipping one preference label induces a parameter-independent shift in the DPO gradient. Using this key property, we can then convert the targeted poisoning problem into a structured binary sparse approximation problem. To solve this problem, we develop two attack methods: Binary-Aware Lattice Attack (BAL-A) and Binary Matching Pursuit Attack (BMP-A). BAL-A embeds the binary flip selection problem into a binary-aware lattice and applies Lenstra-Lenstra-Lovász reduction and Babai's nearest plane algorithm; we provide sufficient conditions that enforce binary coefficients and recover the minimum-flip objective. BMP-A adapts binary matching pursuit to our non-normalized gradient dictionary and yields coherence-based recovery guarantees and robustness (impossibility) certificates for $K$-flip budgets. Experiments on synthetic dictionaries and the Stanford Human Preferences dataset validate the theory and highlight how dictionary geometry governs attack success.

We study the problem of black-box optimization of a function f of any dimension, given function evaluations perturbed by noise. The function is assumed to be locally smooth around one of its global optima, but this smoothness is unknown. Our contribution is an adaptive optimization algorithm, POO or parallel optimistic optimization, that is able to deal with this setting. POO performs almost as well as the best known algorithms requiring the knowledge of the smoothness. Furthermore, POO works for a larger class of functions than what was previously considered, especially for functions that are difficult to optimize, in a very precise sense. We provide a finite-time analysis of POO's performance, which shows that its error after n evaluations is at most a factor of sqrt(ln n) away from the error of the best known optimization algorithms using the knowledge of the smoothness.

Middle-mile logistics describes the problem of routing parcels through a network of hubs linked by trucks with finite capacity. We rephrase this as a multi-object goal-conditioned MDP. Our method combines graph neural networks with model-free RL, extracting small feature graphs from the environment state.

In this work, we formulate a new multi-task active learning setting in which the learner's goal is to solve multiple matrix completion problems simultaneously. At each round, the learner can choose from which matrix it receives a sample from an entry drawn uniformly at random. Our main practical motivation is market segmentation, where the matrices represent different regions with different preferences of the customers. The challenge in this setting is that each of the matrices can be of a different size and also of a different rank which is unknown. We provide and analyze a new algorithm, MAlocate that is able to adapt to the unknown ranks of the different matrices. We then give a lower-bound showing that our strategy is minimax-optimal and demonstrate its performance with synthetic experiments.

We present a machine learning framework for testing general relativity (GR) with gravitational wave signals from binary black hole mergers. Using the source parameters of 173 BBH events from the GWTC catalog as a realistic astrophysical population, we generate simulated GR waveforms and construct beyond GR (BGR) waveforms by applying controlled phase deformations. We introduce a response function formalism that provides a systematic framework for quantifying how any observable responds to modifications of GR. We train convolutional neural networks (CNNs) on two input representations: whitened waveforms and a response function type observable derived from the waveform mismatch, which isolates the effect of phase deviations from the bulk signal. Using response functions as the CNN input improves the classification sensitivity by a factor of approximately 33 compared to whitened waveforms, demonstrating that the choice of observable representation is as important as the classifier architecture. We study the fundamental limits of this classification through Bayes optimal error analysis, averaging methods that reveal coherent patterns hidden in noise, and a comparison between CNN accuracy and a single feature classifier as a proxy for human performance. At all deformation scales, the CNN outperforms the best single feature approach. We extend the framework to physically motivated theories using the parameterized post Einsteinian (ppE) formalism and apply it to massive gravity, where the classifier detects deviations for graviton masses of order $m_g \sim 10^{-23}\;\mathrm{eV}/c^2$ with aLIGO design sensitivity.

We study estimation and clustering in Gaussian mixture models under variance misspecification. Observations are generated with true variance $\sigma^2$, while the component means are estimated using a likelihood with variance $\tau^2$, yielding a family of mismatched likelihood functions parameterized by the ratio $\rho=\tau/\sigma$. We show that the interplay between $\rho$ and the signal-to-noise ratio (SNR) induces a sharp phase diagram. Under correct specification ($\rho=1$), maximum likelihood recovers the true means, independently of the SNR. However, once the model is misspecified, two different regimes emerge. Under under-smoothing ($\rho<1$), the estimated Gaussian means are displaced from the truth, and in low SNR this discrepancy grows as the SNR decreases: for every fixed $\rho<1$, the squared error scales as $\mathrm{SNR}^{-1}$. Under over-smoothing ($\rho>1$), the fitted likelihood blurs the cluster separation, causing distinct component means to collapse towards the overall mixture center once $\rho^2$ exceeds a threshold of the form $1 + \lambda\,\mathrm{SNR}$, where $\lambda$ depends on the geometry of the true means. We further show that the hard assignment objective arises as the limit $\tau\to 0$ of the same mismatched likelihood family, and derive corresponding low- and high-SNR results for hard-assignment mean estimation and latent-label recovery. Furthermore, in low SNR, Bayes-optimal clustering is close to random guessing, and the hard-assignment target remains far from the true means. These results show that in low-SNR applications, even mild variance misspecification or hard-assignment procedures can induce substantial bias, whereas in high SNR these effects are largely absent.

Models for cross-sectional network data have become increasingly well-developed in recent decades, and are widely used. This has led to a growing interest in the connection between such cross-sectional models and the behavioral processes from which the corresponding networks were presumably generated. Here, we build on prior work in this area to present a behavioral micro-foundation for cross-sectional network models, based on a continuous time stochastic choice mechanism, that can accommodate highly general classes of cases (including agents who are not themselves in the network, and multilateral edge control). As we show, the equilibrium behavior of this process under appropriate conditions can be expressed in exponential family form, allowing estimation of individual preferences using existing methods; the graph potential separates naturally into a preference-based term reflecting agent utilities, and an entropic term reflecting the rules of tie formation. We illustrate our approach via an analysis of friendship in a professional organization, and modeling of phase transitions in the structure of small groups.

The Distributional Alignment Game framework provides a powerful variational perspective on Answer-Level Fine-Tuning (ALFT). However, standard algorithms for these games rely on estimating logarithmic rewards from small batches, introducing a systematic bias due to Jensen's inequality that can destabilize training. In this paper, we systematically resolve this structural estimation bias. First, we generalize the alignment game to arbitrary Bregman divergences, showing that for a family of geometries inducing polynomial rewards, we can construct provably exact and unbiased estimators using U-statistics. Second, for the canonical KL divergence game where an exact solution is impossible, we derive a globally robust minimax polynomial estimator that is provably optimal, achieving the fundamental statistical error limit of $\Theta(1/K^2)$, which we establish via the Ditzian-Totik theorem. Finally, we synthesize these two approaches to propose a novel Variance-Optimal Augmented Polynomial Optimization Program (AQP) Estimator, proving that by systematically reducing variance, our method achieves not only optimal bias but also provably accelerated game convergence, leading to more efficient and stable training with zero online computational overhead.

This paper studies sample-size design for finite-population test-and-roll experiments, where a decision-maker first conducts an experiment on $m$ units and then assigns the remaining $N-m$ units to the treatment that performs better in the experiment. We consider welfare-aware sample-size choice, which involves an exploration-exploitation tradeoff: larger experiments improve the rollout decision but impose welfare losses on experimental units assigned to the inferior treatment. We show that the standard absolute minimax regret criterion can lead to implausibly small experiments by over-penalizing exploration in its worst-case objective. To address this limitation, we propose the Worst-case Marginal Benefit (WMB) rule, which compares the worst-case marginal benefit of adding one more matched pair to the experiment with the corresponding marginal exploration cost. We establish a simple rule-of-thirds benchmark. For Bernoulli outcomes, after excluding pathological cases, the WMB criterion yields the optimal sample size of $m \approx N/3$ through a Gaussian approximation. For Gaussian outcomes with a known common variance, the same benchmark arises exactly. These results provide a prior-free and practically implementable guide for welfare-based sample-size design.

Introduction: Normalised prediction distribution errors (npde) are used to graphically and statistically evaluate continuous responses in non-linear mixed effect models. Here, our aim was to extend npde for categorical data and to evaluate their performance. We applied our approach to a real case-study describing the evolution of severe onychomycosis (toenail infection) in a trial comparing two treatment groups. Methods: Let V denote a dataset with categorical observations. The null hypothesis H0 is that observations in V can be described by a model M. Residuals called npde can be adapted to categorical observations using jittering techniques. Their theoretical standard normal distribution can be evaluated through the Kolmogorov-Smirnov test. We evaluated the performance in terms of power through a simulation and compared it to a Chi-square. We illustrated the test and graphs on a real case-study. Results: npd were able to detect misspecifications in the structural model and model parameter value. As expected, the power to detect model misspecifications increased both with the difference in the shape of the probability, and with the sample size. Chi-square test performed better but npd could be readily applied in all type of design. Based on the toe-nail data, graphs reveal a huge discrepancy of the base model, and a good adequation for the best model we found. Conclusions: npde can be extended to categorical data, particularly in clinical settings with unbalanced design and graphs can be useful to evaluate the model as well as the covariate effects.

Introduction: Joint models are increasingly used in clinical trials. An important part of model building is to properly assess the descriptive and predictive ability of these models. Normalised prediction discrepancies (npd) and normalised prediction distribution errors (npde) have been developed to evaluate graphically and statistically non-linear mixed effect models for continuous responses. In this work, we propose to use a combined test to evaluate joint models. Methods: Prediction discrepancies (pd) are defined as the quantile of the observation within its predictive distribution and obtained by Monte-Carlo simulations. The pd for unobserved (censored) event times are imputed in a uniform distribution based on the model prediction of the probability of censoring, using a similar method as the one developed to handle data under the lower quantification limit (LOQ). We propose to combine the p-values of the tests on longitudinal data and on time-to-event (TTE) data, adjusted with a Bonferroni correction. We performed simulation studies based on a joint model characterising the relationship between prostate specific antigen biomarker (PSA) and survival in prostate cancer patients to evaluate the type I error and power of npd/npde to detect different types of model misspecifications. Results: For all types of misspecifications, the type I error of the combined test was found to be close to the expected 5%. The power of the combined test to detect model misspecifications increased with the difference from the true model and as expected, with sample size. Graphically the power increase can be related to larger differences in the shape of the survival function or PSA evolution. Conclusions: npd can be readily extended for event data by imputing the pd for censored event under the model. The test showed an adequate type I error, and was quite sensitive to alternative models tested.