Papers

Group-based reinforcement learning (RL) methods, such as GRPO and its variants, have become a leading paradigm for training reasoning and agentic large language models (LLMs). While their group-normalized advantage estimation is reliable at the response level, it becomes systematically biased at the step level, since coarse-grained trajectory-level advantages are hard to accurately reflect the contribution of individual steps (i.e, failed trajectories may contain valuable steps). Revisiting the foundational RL definition, we notice that GRPO's success on single-turn tasks stems from its advantage estimation strategy, which adheres to the basic definition: the mean reward of multiple actions sampled from the same state constitutes a credible state-value estimate. Extending the faithful estimation to step-level would in principle demand sampling multiple actions from each intermediate state, which is too costly on a per-state basis. To mitigate this issue, we propose a Graph-based Faithful sTep-level credit-assignment framework (GRAFT) that grafts all rollout trajectories into a trajectory graph, recovering node state-values via Bellman iteration on the graph, and assigning credit to each edge by the node value difference. Theoretically, the estimated step-level advantage faithfully adheres to the basic advantage definition in RL. To further ensure the reliability of step-level advantage estimation, we further propose Graph GAE, which extends GAE to the trajectory graph for reducing the impact of state-value estimation bias. Experiments across a range of multi-turn agentic benchmarks show consistent gains over GRPO and superior performance compared to recent agentic RL algorithms. Code will be available at https://github.com/xcyao00/GRAFT.

Reliable intervention-oriented uncertainty quantification from Earth observation (EO) remains challenging when continuous treatment shifts, spatial dependence, limited support, and satellite-outcome uncertainty must be addressed simultaneously. Existing causal, conformal, and spatial approaches address parts of this problem, but their direct combination does not generally recover the appropriate interventional reference law because candidate reassignment jointly alters treatment likelihood, standardized residuals, and graph-dependent residual likelihood. This study presents GeoDose-CP, a support-aware conformal framework for localized stochastic potential outcomes under continuous or mixed continuous-atomic treatment. Its central methodological contribution is a graph-local target-orbit law that jointly represents intervention-induced treatment shift, the inverse outcome-scale Jacobian, and spatial residual dependence. The framework further provides exact weighted candidate inversion, a scalable sparse approximation with explicit discrepancy accounting, and refusal under inadequate support. Evaluation used controlled known-truth experiments, MineDoseBench, treatment-density sensitivity analysis, external conformal comparators, and a multi-mine New South Wales (NSW) study. In MineDoseBench, GeoDose-CP achieved mean selective coverage of 0.9692 across 27 configurations and a minimum local q0.05 of 0.8951; exact-sparse auditing produced nine inclusion disagreements over 2,700 targets. In the NSW study, the absence of an auditable longitudinal rehabilitation treatment rendered treatment-dependent inference nonoperational rather than forcing inference through a proxy exposure.

We establish moment and mixed-tail inequalities for fixed-order decoupled homogeneous chaoses generated by independent, centered, sparse bounded random variables. Our bounds apply to arbitrary real rectangular coefficient tensors and describe the fluctuation scales through weighted slice and partition norms, with a Bennett-type logarithmic improvement in the largest-entry term. As an application, we derive guarantees for sparse Khatri–Rao embeddings that explicitly account for sparsity and input geometry.

Covariates summarized over a subject's completed follow-up are sometimes entered into Cox regression as though observed at baseline. This practice incorporates future event or censoring information and changes both the estimand and its sampling behavior. We analyze an affine class in which a genuine baseline covariate is contaminated by realized follow-up time. Treating partial likelihood as an observed-data estimation criterion, we derive the population score and characterize its unique least-false coefficient. Exact continuous-time benchmarks show scale reduction and saturation under strong contamination, while administrative censoring destroys the reduction and may produce overshoot. Grouping exit times with Breslow ties changes the geometry: the coefficient has a single hump and eventually returns to zero even though the induced association diverges. We also derive an observed-data influence function and show why model-based variance can be either too small or too large. A subject-level sandwich consistently estimates uncertainty around the least-false target under the stated conditions, but it does not correct the target itself.

Panel conditioning, the causal effect of prior survey participation on responses, can vary with tenure. Under an additive model of cell means in period, entry cohort, and tenure, we characterize which features of the conditioning path a staggered panel identifies on its observed support, and how the unidentified component affects common panel estimators. The identified set of the path is an affine translate of the tenure projection of the cell design's kernel, and a linear functional of the path is identified exactly when it annihilates that projection. It always contains an affine direction and, when the entry cohorts share a stride, periodic directions, which exhaust it under a connectivity condition on observed increments; second differences at that stride are then identified, and ordinary ones generally are not when the stride exceeds one. Under a recruitment condition, an interrupted schedule such as the four-eight-four rotation of the Current Population Survey (CPS) distinguishes a constant increment per interview from one per calendar month, which no equally spaced schedule can. We give support conditions for recovery under a plateau, entry-wave negative controls, or bounded cohort drift. A second set of results links identification to regression: two-way fixed effects absorb every unidentified direction, so the remaining conditioning bias is normalization-invariant and itself identified, and a two-way regression with tenure indicators corrects it under a residual-rank condition. When event time is aligned with tenure, conditioning shifts event-study coefficients by a known linear functional of the path, producing pre-trends without anticipation; bounds on identified curvature bound those shifts. Simulations verify the identities, a 19-wave Japanese panel illustrates the support calculations, and published CPS month-in-sample indices give a descriptive, not identifying, example.

Competing-risks data do not identify latent marginal duration distributions or their dependence without additional restrictions. This paper asks whether restrictions introduced to restore identification themselves restrict the observable law. For a two-risk Archimedean model with categorical exclusion restrictions, we derive a necessary-and-sufficient observable characterization. A discrete single-crossing argument identifies the scalar copula parameter from cell-specific overall survival probabilities, while cause indicators recover the remaining allocation and generate additional specification restrictions. We construct an identification-robust, self-normalized quadratic statistic, invert it to obtain confidence sets, and use empty inverted sets as a conservative specification test. Simulations show that the cause indicator can be decisive: under the weakest contrast considered it converts a frequently uninformative survival-based confidence set into an informative joint set without loss of coverage, whereas excessive categorical contrast can eliminate causes from individual cells and make cause-specific recovery inadmissible. The results turn latent exclusion restrictions into refutable restrictions without estimating covariate derivatives.

Large language models are increasingly used as inexpensive judges to evaluate outputs, label data, and assess whether a system meets a desired quality standard. Yet using AI judgments for formal statistical inference is fundamentally different from simply treating them as ground-truth labels: AI evaluations can be biased or noisy, and rigorous hypothesis testing requires explicit control of type-I and type-II errors. We study how to use AI judgments, together with selective human verification, to conduct a valid hypothesis test at minimum cost. We consider a population of items with hidden binary labels. After choosing a fixed pool of items, the decision maker can selectively query AI, send an item directly to a human, escalate an AI-scored item to a human after observing the AI report, or stop once sufficient evidence has accumulated. We derive an information-theoretic lower bound that captures the minimum cost of achieving prescribed testing errors and characterizes the value of AI information and human verification through a report-dependent information frontier. Motivated by this characterization, we develop SCALE, a sequential cost-aware policy that combines selective AI scoring with adaptive human escalation. SCALE is valid at finite sample sizes and matches the lower bound to first order as the target error probabilities vanish. We further extend the framework to an unknown AI-output model using paired AI-human pilot data. Numerically, SCALE approaches Human-only or AI-only testing when one source clearly dominates, while achieving its largest savings when inexpensive AI judgments and selective human verification are both valuable.

Mixture-of-Experts (MoE) models have emerged as an effective paradigm for scaling large language models, yet their massive parameter footprint poses severe memory challenges for deployment. Existing compression methods typically treat expert pruning and merging as disjoint operations and rely on greedy heuristics, often leading to irreversible information loss or representational pollution. In this paper, we propose PRIME, a unified optimization framework for MoE compression via adaptive PRunIng and MErging. PRIME casts the discrete prune-merge assignment step as a global Integer Linear Programming over a frequency-weighted surrogate objective, jointly selecting pruning and merging decisions while accounting for expert usage. We motivate the frequency-aware objective through an error-bound analysis under fixed representative assignments, and show that the formulation favors high-frequency experts as centroids, promoting routing stability. Extensive experiments on Qwen3-30B-A3B and Qwen3.5-122B-A10B demonstrate that PRIME achieves strong accuracy retention across knowledge, coding, math, and long-context tasks, while serving benchmarks of merged models quantify the memory and throughput benefits of expert compression in vLLM.

Adaptive arm selection changes the distribution of the observations collected by a bandit algorithm, but it need not change their limiting empirical spectrum. We study Gaussian bandit designs in which the dimension and the number of observations grow proportionally. A quantitative coupling theorem compares the design generated by any causal selection rule with an independent Gaussian design. If the logarithm of the number of available arms is sublinear in the dimension, the empirical spectral distribution converges to the Marchenko-Pastur law, uniformly over the selection rule. Consequently, Gaussian Bayesian bandits have policy-independent first-order limits for posterior mean-square uncertainty, squared posterior covariance, and information acquisition. For linear-score selection, we obtain the exact conditional arm distribution and show that two-arm selection produces an exactly Wishart Gram matrix in every dimension, despite its nonzero conditional mean. For a fixed selection direction, we identify an explicit eigenvalue and eigenvector transition governed by the second moment of a Gaussian maximum. A counterexample shows that a direction's overlap with a reference signal does not determine this transition. Finally, an exponentially large arm pool permits a different bulk limit, establishing the order-sharpness of the arm-growth condition. These results distinguish global spectral stability from directional effects in adaptive bandit data.

Causal discovery from observational and interventional data becomes challenging in the presence of latent confounding and selection bias, where causal structure is no longer adequately represented by directed acyclic graphs over observed variables. Existing model-free methods often rely on an exponential number of conditional independence tests and provide limited uncertainty quantification in high-dimensional settings. We develop a model-free and constraint-query optimal statistical inference framework for causal discovery under latent variables and selection using single-target interventions. We introduce the system-induced subgraph (SIS) to capture the causal relations among system variables while accounting for context variables. We establish its identifiability through maximal ancestral graphs (MAGs), and show that interventions on each observed system variable are sufficient for unique identification and necessary in the worst case. Building on these results, we develop a two-stage graph inference procedure with asymptotic family-wise error control under sufficient first-stage power. For $d_X$ observed system variables, the procedure requires at most $\frac{5}{2}d_X^2$ statistical tests, parallelizable within each stage, and achieves optimal constraint-query complexity up to a constant factor. The framework accommodates soft interventions and avoids parametric structural equation assumptions. We illustrate the methods through analysis of Perturb-seq data from interferon-$\beta$-stimulated A549 lung cancer cell lines.

Matrix time series is a series of matrix data observed over time. Analytical tools for such time series is needed in many applications in finance, economics, engineering and many other fields. To avoid the use of vectorization of the matrices which loses the column and row information, and the vector autoregression framework in traditional time series analysis, \cite{chen2021autoregressive} proposed the Matrix Autoregressive (MAR) Model. The model maintains and utilizes the matrix structure, leading to a substantial dimensional reduction and admitting explicit interpretations, comparing with the vector autoregressive model on the vectorized data. However, the MAR model still encounters difficulties in dealing with large dimensional matrix time series as the coefficient matrices in MAR models are also large. In this paper we propose to achieve further dimension reduction through reduced-rank constraints of the coefficient matrices in the MAR model. Estimation and rank determination procedures are studied. Theoretical investigation and empirical examples show that the reduced-rank constraint can achieve higher statistical efficiency than the MAR model.

We introduce GRACE (the General Regime-Aware Changepoint Estimator), a general, globally exact multiple-changepoint detection and estimation framework in which arbitrary numbers of model parameters may change at or persist across estimated changepoint locations, and all permissible changepoint types and patterns of parameter sharing compete within a single $\ell_0$-penalized optimization. The analyst specifies which types of changepoints are permitted and which parameters each type must preserve, without specifying which changes actually occur, where they occur, or in what order. GRACE jointly estimates the number and locations of changepoints, the type of change at each location, the model fitted between changepoints, and all continuous parameters. As a direct consequence, GRACE can estimate models in which the slope persists across a mean shift. The generality of the framework is achieved through an exact dynamic-programming recursion that retains the accumulated fitting cost conditional on whichever parameters a subsequent change may preserve, tracking multiple parameters jointly when necessary. We establish the global exactness of the recursion and develop functional, penalty-based, and objective-bound pruning rules that eliminate candidates without sacrificing optimality. Ordinary optimal partitioning, FPOP, and CPOP are recovered as special cases, while piecewise-polynomial and seasonal models illustrate the broader range of permissible segment structures and transitions. The resulting framework supports heterogeneous changepoint types, parameter sharing across segments, and substantial parallel computation within a unified approach to exact changepoint estimation.

Estimating causal effects is challenging with observational data due to the lack of exchangeability, and is further complicated with spatial data. In real-life applications such as agricultural and environmental studies, unobserved spatial factors (spatial confounding, SC) and interactions of units in nearby locations (spatial interference, SI) commonly occur jointly, yet existing parametric methods address them separately. We develop a unified framework that treats both simultaneously. We first define the estimand precisely and give an identification result. We then study two estimators. The matching estimator iDAPS integrates neighbourhood exposure and spatial proximity into propensity score matching with balance-optimised weights. We prove a variogram bias bound showing its confounding bias is controlled by the spatial variability of the confounder at the matched distance; this justifies the composite metric, yields a consistency result, and provides a computable diagnostic. The doubly robust estimator recoverU+ augments the propensity and outcome models with the recovered confounder and the exposure; we prove it is doubly robust up to an explicit residual bias from the unrecoverable confounder component. We show that independence-based standard errors are anticonservative under spatial dependence; we provide spatial heteroskedasticity and autocorrelation consistent (HAC), block-bootstrap, and randomisation-based alternatives whose calibration we verify. Every theoretical result is validated numerically. Estimating the effect of selective catalytic and non-catalytic reduction (SCR/SNCR) technologies on ambient ozone reveals no evidence of ozone reduction, while a naive analysis would have delivered a confidently wrong sign. All methods are implemented in the developed open-source R package spaci.

We initiate the study of approximating the top eigenvalue and eigenvector of a random symmetric matrix $ A \in \mathbb{R}^{n\times n} $ using $ q(A)b $ where $q$ is a degree-$d$ polynomial and $b$ is a standard Gaussian vector independent of $A$. For spiked GOE $ Y = \lambda vv^\top + X $, we identify $ d_\star = \frac{\log(n)}{2\log(\lambda)} $ to be the critical degree threshold above which accurate approximation of the top eigenvalue and eigenvector is possible. This sharpens the common belief that spectral methods can be implemented by $ O(\log(n)) $-step power iterations and offers a precise connection between spectral methods and low-degree polynomial algorithms, a popular proxy for all polynomial-time algorithms. For GOE $X$, we identify $ d_\star = n^{1/3+o(1)} $ to be the critical degree threshold for top eigenvector approximation, whereas constant degree suffices for top eigenvalue approximation. Moreover, in the limit where $ d/n^{1/3} $ converges to a positive finite constant, we compute the exact asymptotic eigenvector approximation accuracy in terms of the expected squared overlap. These results significantly improve upon predictions made in randomized numerical linear algebra for deterministic data matrices that the iteration count of power methods with random initialization is governed by the inverse spectral gap. Technically, our analyses leverage extremal properties of Chebyshev polynomials and draw upon the rich literature of random matrix theory.

This note presents a general form for a proper scoring rule that can be used for possibly right-censored survival data without requiring any knowledge or assumptions about the censoring process.

Deep clustering is a powerful approach for discovering meaningful structures in high-dimensional data by learning a low-dimensional latent representation prior to clustering. Despite its empirical success, assessing the statistical reliability of the resulting clusters remains challenging. Testing discovered clusters on the same data induces selection bias and invalidates classical $p$-values. Selective inference (SI) provides a principled framework for correcting this bias, but existing methods focus on clustering performed directly on the observed features. In this work, we develop an SI framework for deep clustering with a fixed pretrained encoder. The key challenge is that cluster assignments are determined through a nonlinear transformation from the original data space to the latent space, resulting in a substantially more complex selection process than in conventional clustering. Our method provides a computationally tractable way to account for this process and enables valid statistical testing of differences between clusters identified in the latent space. Synthetic experiments demonstrate that the proposed method controls the Type I error rate while achieving higher power than valid but conservative baselines, and genomic applications show that it can identify significant cluster differences while appropriately accounting for selection bias. Our framework provides a principled approach to quantifying the statistical reliability of structures discovered by deep clustering.

We study Bayesian linear bandits with an isotropic Gaussian parameter, independent Gaussian candidate arms, and Gaussian reward noise when the horizon is proportional to the dimension. The normalized posterior uncertainty has an explicit limit that is uniform over all causal policies. Gaussian posterior identities then determine the limiting parameter overlaps without an assumed closure of the adaptive recursion. These results yield exact regret curves for Thompson sampling, posterior-mean greedy selection, and a family of policies that scale the posterior sampling covariance. The normalized realized cumulative regret converges in L1, uniformly on compact proportional-time intervals. A policy-uniform lower bound identifies the limiting optimal Bayes regret and proves that posterior-mean greedy selection attains it. Thompson sampling incurs a strictly larger leading regret; its instantaneous regret ratio relative to greedy selection lies between one and two and approaches two at long proportional horizons. Closed-form cumulative curves also identify a different comparison in the vanishing-noise limit. Finally, the instantaneous regret converges to a nondegenerate Gaussian decision-loss distribution, rather than to its mean. The analysis separates the amount of information acquired by a bandit policy from the quality of the decisions made using that information.

In this paper, one provides a comprehensive mathematical and practical overview of Functional Data Analysis (FDA) specifically applied to parametrized curves in Rp. One observes that curves depending on continuously parameter are naturally present across many fields, such as, for example, monitoring child development in pediatrics, assessing meteorological phenomena, analyzing financial portfolios, tracking neurological functions, and mapping geographic pollution levels. One considers a formal theoretical framework given by a Cartesian product of real separable Hilbert spaces, to model these curves mathematically. To bridge the gap between discrete experimental measurements subject to errors, and smooth continuous functions, the paper details the essential phase of data smoothing and fitting and sets its mathematical formulation via of the Ordinary or Penalized Least Squares criteria which, for the later, using the Sobolev spaces framework, incorporates a smoothing parameter ${\lambda}$ and differential operators to prevent erratic geometric behaviors by penalizing excessive curve roughness. Furthermore, one extends the usual Principal Component Analysis to its functional counterpart in $\mathbb{R}^3$, using the calculus of variations and the Euler-Lagrange multiplier theorem in order to find eigenfunctions and eigenvalues of the covariance operator to exhibit the optimal decomposition of the spatial variance and finally, one highlights the geometric advantages of FDA over traditional multivariate data analysis, emphasizing its unique capacity to capture critical differential features as velocity, acceleration, curvature, and torsion.

Despite the importance of reinforcement learning with verifiable rewards (RLVR), the extent to which it can learn new reasoning capabilities remains debated. Here we study the optimization landscape of RLVR on algorithmic tasks, such as iterated group and quasigroup multiplication. To this end, we map entropy-regularized RLVR over myopic tabular policies onto an energy-based (spin-glass) model over deterministic policies. This mapping upper-bounds what RLVR can achieve, and lets us rigorously characterize the landscape in this tabular setting. We show, both theoretically and experimentally, that for a wide class of models and tasks with uncorrelated inputs, this landscape is benign, containing no local minima that could trap RLVR training. Rather, the practical difficulty of these tasks appears to stem, at least in part, from issues such as diffusive barriers and gradient-estimation error in traversing the landscape. These are genuine obstacles that can prevent a solution from being found, but they are distinct from the landscape itself being rugged. We show that these obstacles can often be mitigated through the choice of entropy regulator. Consistent with this theory, we find that a transformer trained from scratch, using only last-token rewards, successfully learns an algorithmic chain of thought for iterated non-Abelian group multiplications.

We study global convergence guarantees of third-order Langevin dynamics for non-convex optimization via simulated annealing with fixed friction and decreasing noise. An explicit three-block distorted entropy transfers dissipation from the noisy auxiliary variable to the full state. Under dissipativity, regularity, and low-temperature functional-inequality assumptions, logarithmic cooling drives the objective values to the global minimum in probability at the barrier-controlled kinetic rate. For the exact-force-integral and midpoint three-stage discretizations, polynomially decreasing steps preserve this rate on the physical time scale. The cubic local endpoint estimate gives a less restrictive sufficient step-size condition than the available frozen-force kinetic result. A comparison with the one-gradient UBU integrator shows how its centered stochastic local error leads, under the same strong-coupling analysis, to a smaller sufficient iteration exponent. Numerical experiments are conducted to illustrate our theory. For a double well objective, third-order Langevin terminal-success point estimates are higher than UBU at both a common horizon and an equal gradient budget. For a high-dimensional nonconvex neural-network objective using synthetic data, independently tuned UBU and third-order Langevin schemes both outperform overdamped Langevin dynamics; the third-order Langevin point estimate is higher. For the same neural-network objective on real data, we show the same point-estimate ordering for best-basin probability and post-quench test accuracy. Numerical code and associated experiment results are publicly available at https://github.com/gagawjbytw/simulated-annealing-third-order-langevin.

Large language models can interpret natural lan- guage, yet robust decisions remain challenging. Jev-like models expose structured choices, but these interfaces do not directly provide numeri- cal values at a requested precision. We propose NUMERICJEV, a training-free numerical decod- ing algorithm that enables numerical output from any LLM with a Jev-like structured-choice in- terface. Surprisingly, on our arithmetic bench- mark, it outperforms direct selection from a can- didate list containing the correct answer by 2.93 percentage points (Figure 1). Our motivation comes from the observation that numerical range selection is itself a decision problem that Jev- like LLMs can address. NUMERICJEV recur- sively refines a range through a multiway deci- sion tree while retaining the original question in context, without parameter updates or hidden- state access. On a 100-value grid, a ten-way tree requires only two decision rounds. Range- normalized MAE is 1.84% versus 5.18% for di- rect choice. A separate three-date historical- index study yields 4.58% mean relative recall er- ror and 0% readout error when the value is sup- plied. Code is available at https://github. com/Bring-AI/jev-numeric.

The recent emergence of Time Series Foundation Models (TSFMs) has significantly advanced multi-step forecasting performance, enabling accurate predictions over extended future horizons. However, existing TSFMs often suffer from significantly inherent uncertainty, which typically manifests as derived forecast branches emerging at each time step and spreading to subsequent steps; different forecast branches often exhibit varying forecasting performance, thereby undermining the credibility of TSFM forecasts. In this paper, we propose the Slicing-Graphing-Alignment (SGA) method to quantify the uncertainty of multi-step TSFM forecasts. The proposed SGA first characterizes the topology of all potential forecast branches using a directed acyclic graph, such that the graph complexity bounds the uncertainty of multi-step forecasts, and then precisely measures the graph complexity by integrating both topological information and TSFM-inherent stochasticity. Experimental results conducted on 11 TSFMs and 27 datasets demonstrate that (i) SGA achieves the best performance when ranking predictive errors with uncertainty estimates; (ii) SGA works with a more extensive and more precise sampling coverage than those of existing UQ methods, deriving a quantification mechanism fundamentally different from those of established ones; and (iii) larger model scales of TSFMs correlate with lower uncertainty estimates of multi-step forecasts, suggesting another empirical scaling law for uncertainty quantification of multi-step TSFM forecasts.

Aggregation theory has traditionally focused on operators defined over vectors. However, many applications-including Multi-Criteria Decision Making, Group Decision Making, Fuzzy Rule-Based Classification Systems, and overlap/grouping indices-require aggregating information naturally structured as a matrix of membership degrees (e.g., where a set of objects interacts with a family of fuzzy sets). Despite this, no formal framework has been proposed for this class of operators, partly due to the common practice of flattening matrices into vectors (which discards structural information) and partly due to a reliance on decomposable operators that aggregate rows and columns sequentially. This paper addresses this gap by formalizing the notion of a matrix aggregation operator (MAO). We analyze the decomposability and symmetry properties of MAOs, showing that certain operators cannot be expressed in decomposable form and examining several notions of symmetry. Finally, we introduce a family of MAOs termed maximum entropy global coverage indices (MEGCIs), provide a construction method for them based on combining grouping functions and MEOWA operators, and illustrate their usefulness in cluster quality assessment through an extensive computational study.

Evaluating a stochastic large language model is costly: benchmark scores estimate expected performance from randomized rollouts, yet uniform repetition ignores sharp differences in task-level rollout variance. We ask how to minimize the variance of a fixed-benchmark mean under an exact rollout budget. We develop Speculative Evaluation with a Hierarchical Bayesian Neyman (HBN) policy with pilot size and stage weight jointly chosen ex ante. It runs a short uniform pilot, pools per-task success counts with a hierarchical Bayesian model, and uses posterior expectations of task-level sampling variances for exact positive-integer Neyman allocation. To mitigate the pilot synchronization barrier, HBN-async speculatively executes continuations from partial pilot feedback and retains those selected by the final allocation. Across six checkpoints and 18 benchmark groups, we evaluate 107 nondegenerate benchmark-checkpoint profiles. For rollout budgets of 8-64 per task, Speculative Evaluation reduces variance relative to Uniform by 12.8%-33.6% on average across profiles, outperforming hindsight-tuned empirical and independent Bayesian baselines. Real-generation experiments that account for the pilot synchronization barrier show that HBN-async mitigates its overhead, helping translate statistical efficiency into practical evaluation benefits.

When can a learner make only finitely many prediction errors along every infinite sequence labeled by a fixed, unknown hypothesis? We characterize this form of consistency for arbitrary binary hypothesis classes in ZFC, without requiring a uniform mistake bound. The characterization uses a single linear order on finite realizable traces. Each trace selects its least subtrace, and the order must satisfy two conditions: conflicting traces select different subtraces, and the order is well-founded on the traces of each fixed target. These conditions induce a learner whose selected evidence decreases on every mistake. Conversely, a consistent learner yields such an order through canonical mistake transcripts and the Kleene–Brouwer ordering. The result provides a representation of consistent prediction by finite evidence, answering a question of Lu (2024).

The Baringhaus-Henze-Epps-Pulley (BHEP) tests for multivariate normality are affine-invariant goodness-of-fit tests based on a Gaussian-weighted $L^2$ distance between empirical and Gaussian characteristic functions. In 1990, Henze and Zirkler expressed the limiting null distribution through the eigenvalues of an integral operator on the standard Gaussian space. In 1997, Henze and Wagner obtained a simpler covariance kernel and raised the problem of calculating the eigenvalues of the resulting operator on a Gaussian-weighted space. Although subsequent work treated the univariate case and numerical approximations in a few low dimensions, the complete all-dimensional spectral problem remained open. This paper determines both complete spectra for every dimension $d \in \mathbb{N}$ and every smoothing parameter $\beta > 0$. The two operators are shown to have the forms $\mathcal{X}_{\beta,d}^*\mathcal{X}_{\beta,d}$ and $\mathcal{X}_{\beta,d}\mathcal{X}_{\beta,d}^*$ for the same Hilbert-Schmidt operator $\mathcal{X}_{\beta,d}$. Consequently, their nonzero eigenvalues agree, including multiplicities, while the null space of the Henze-Zirkler operator is identified exactly. The Gaussian integral operator in the Henze-Wagner decomposition is diagonalized by Mehler's formula, and rotational symmetry confines the finite-rank correction to the sectors associated with spherical harmonics of degrees $0$, $1$, and $2$. The degree-$1$ and degree-$2$ eigenvalues are characterized by scalar transcendental equations, and the radial eigenvalues by an explicit pole-safe Fredholm determinant. The paper establishes nonnegativity, multiplicities, eigenfunction reconstruction, completeness, the trace identity, and a complete characterization of all exceptional pole cases.

We study structural-change testing in multivariate linear regression when the response dimension is proportional to the sample size and the number of predictors is fixed. Despite its relevance to applications across a broad range of fields, this problem remains underexplored. The alternatives of interest allow multiple unknown changes in a prescribed linear contrast of the predictor effects. We construct a least-squares-based Wald statistic standardized by the residual covariance estimator and scan it over candidate change-point segmentations. We propose flexible segment scans for both single and multiple change points, together with a discretized multiscale scan that reduces the computational cost. When the response dimension and sample size diverge proportionally, we establish weak convergence of the normalized statistic process to a centered Gaussian process, yielding implementable critical values for all three scans. We further characterize their asymptotic power under local alternatives, explicitly describing how the signal, design, and high-dimensional aspect ratio determine the limiting power. The finite-sample performance is examined through simulation studies. We apply the proposed methods to filtered and standardized returns from U.S. equity portfolios to investigate structural changes in their exposures to the Fama–French factors.

We introduce Sharper Transductive Local Complexity (STLC), a localized complexity method for transductive learning under uniform sampling without replacement. The construction starts from a Bernstein-type concentration inequality for the supremum of the test–train empirical process. Its proof uses the modified log-Sobolev inequality for the swap walk and a two-parameter entropy closure. A peeling argument with a surrogate localization functional then gives excess-risk bounds with the same fixed-point and confidence terms as the classical inductive local Rademacher-complexity bounds, without the additional logarithmic confidence factor in earlier transductive results. For realizable learning over a binary class of VC dimension $\dVC$, with training size $m$, test size $u$, and $u\ge m\ge\dVC$, STLC yields $\cO\{\dVC\log(me/\dVC)/m\}$. This matches the standard inductive rate and, when $m\ge9$, is within a logarithmic factor of the transductive minimax lower bound of order $\dVC/m$. For transductive kernel learning, STLC gives a spectrum-adaptive excess-risk bound without the multiplicative imbalance factors appearing in the earlier local-complexity bound.

Large-scale astronomical surveys provide unprecedented volumes of multivariate time-series observations obtained through multiple optical filters. We develop a structured multivariate continuous-time autoregressive moving average (MCARMA) framework for multi-band time series with irregular sampling, heteroscedastic measurement errors, and partially observed bands. The framework allows band-specific stochastic dynamics while modeling cross-band dependence through correlated Brownian driving processes, with state-space and spectral representations enabling likelihood-based inference and interpretation of the fitted stochastic dynamics. We develop a two-stage estimation procedure in which a numerically stabilized preliminary fit initializes subsequent maximum likelihood estimation. Simulations show that higher-order stochastic structure can be recovered when its characteristic features are adequately resolved, but can become weakly identifiable because of limited temporal resolution or near pole–zero cancellation. Joint multivariate estimation improves parameter recovery in 23 of 27 settings and spectral recovery in 26 of 27 settings relative to separate single-band fits. Three Sloan Digital Sky Survey Stripe 82 quasars, respectively favoring MCARMA(1,0), MCARMA(2,0), and MCARMA(2,1), illustrate how joint multiband modeling uses cross-band dependence to inform marginal dynamics and can yield different model-order and spectral inference. The methodology is implemented in the Python package mcarma.

There is much interest in the potential to use structural health monitoring (SHM) technology to augment traditional nondestructive inspection (NDI) methods to improve safety, increase asset availability, and reduce maintenance and inspection costs. SHM has the potential to be used in many applications, including critical components in aircraft and pipelines. Probability of detection (POD) plays a critical role in aircraft structural integrity programs, leading to increased interest in developing methods to assess POD in SHM applications. In contrast to traditional NDI laboratory experiments involving specimens with cracks, SHM sensors are fixed, and SHM data are acquired over time as cracks grow or otherwise evolve. Thus, traditional statistical methods for assessing POD must be replaced or extended to properly handle repeated-measures data. The purpose of this paper is to review the basic statistical concepts of POD and show how these concepts can be extended or adapted for SHM-POD applications. The paper presents statistical methods for modifying and extending existing POD methods, including a simple size-of-damage-at-detection (SoDaD) method and a random-parameter (RP) method for repeated-measures data. The methods are compared using three case studies involving Piezoelectric Transducer (PZT), Carbon Nanotube (CNT), and Comparative Vacuum Monitoring (CVM) sensor systems. Results show that the SoDaD method provides a simple approach for POD estimation with limited data, while the RP method offers enhanced modeling fidelity by utilizing repeated measurements. These methods are applicable when a scalar damage index or similar response is used to make a detection decision.