PyTorchKR Papers

최신 논문

총 3,947편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 카테고리 구독

arXiv:2607.06652v2cs.LG

From Jumps to Signatures: A Generative Method for Temporal Point Processes

Rough path signatures are a universal feature map for continuous paths and, via the expected signature, characterise path distributions. These guarantees do not directly extend to cadlag paths of Temporal Point Processes (TPPs), limiting the use of signature methods for event sequences. Furthermore, neural TPP models, including recent generative approaches, optimise per-event objectives with no global sequence-level loss, while evaluation of variable-length event sequences lacks distributional discrepancy measures. This paper proposes a common pathwise framework for addressing these limitations. We introduce the interarrival embedding, a stable, injective lift from jump paths to continuous paths of bounded variation, extending signature methods to discrete event sequences. Our theoretical contributions give rise to sigTPP, the first signature-based generative model for TPPs, trained using a path-level loss on complete trajectories. We further analyse the space of counting paths and derive $3$ distributional discrepancies, providing mathematically justified tools for evaluating generative TPP models. Across $12$ tasks, sigTPP achieves the best mean rank based on $8$ complementary metrics, outperforms or is within one standard error of the strongest baseline (which varies based on performance) in $57\%$ of the dataset-metric pairs, and according to a relative score, improves against each baseline by at least $24\%$ on average.

arXiv:2607.02891v2cs.LG

오분류 리덕션을 통한 비정상 선형 밴딧의 동적 후회

Dynamic Regret for Non-Stationary Linear Bandits via Misspecification Reductions

Many online decision-making problems involve both round-specific feasible actions and drifting reward models: eligible ad impressions, feasible prices, and available treatments can change over time, while user preferences, demand curves, and patient responses may evolve. Motivated by these applications, we study non-stationary linear bandits with round-specific feasible decision sets. Our key contribution is a blockwise reduction that connects parameter drift in non-stationary linear bandits to model misspecification in contextual linear bandits while preserving the round-specific decision benchmark. This connection allows us to leverage misspecification-adaptive learning to obtain dynamic-regret guarantees for general time-varying decision sets. For a known path-length budget $P_T$, we establish these guarantees for general compact, adaptive non-anticipating decision sets under an oblivious parameter path, and for $K$-armed contextual linear bandits under an oblivious environment. For general time-varying decision sets, our bounds improve the dependence on $T$ and $P_T$ from the $\widetilde O(T^{3/4}P_T^{1/4})$ guarantees available for existing LinUCB-based approaches to $\widetilde O(T^{2/3}P_T^{1/3})$, matching the non-stationary lower-bound dependence up to logarithmic factors.

arXiv:2606.27090v3stat.ML

전역적 발산을 넘어서: 베이지안 추론에 대한 지역 질량 관점

Beyond Global Divergences: A Local-Mass Perspective on Bayesian Inference

KL 발산(divergence)과 ELBO와 같은 전역적 목적함수(global objective)는 베이지안 추론에서 분포 간 불일치를 측정하기 위해 널리 사용된다. 본 논문은 이러한 전역적 목적함수로는 직접 포착되지 않는 분포의 '지역 질량 행동(local-mass behaviour)'을 연구한다. 우리는 두 가지 수학적 도구를 도입하여 사용한다: (1) 지역 질량의 다항식 및 로그 감쇠 스케일을 기록하기 위한 Mass Index, 그리고 (2) 특이 성분(singular component)이 존재하는 경우에도 정식화할 수 있는 집합 국소화 발산인 정칙화된 확장 KL(regularised extended KL, RE-KL)이다. Mass Index는 베이지안 업데이팅이 지역 질량을 어떻게 변화시키는지 특성화하는 데 도움을 준다: (1) 거듭제곱-로그 가능도(likelihood) 인자는 이를 명시적으로 이동시키며, (2) 매개변수(parameter)에 의존하는 지지집합(support), 또는 그것의 매끄러운 완화(softening)는 매개변수 값 근처에 남아 있는 질량의 양을 통해 지역 스케일을 변화시킬 수 있다. 지역 RE-KL을 사용하여, 우리는 두 가지 KL 방향 하에서 지역 스몰볼(small-ball) 질량을 비교하기 위한 절대적, 상대적, 방향적 부등식을 증명한다. 이러한 결과들은 함께 지역 질량 행동에 대한 지역적인 이론적 설명을 제공한다. 실험은 이러한 지역적 행동에 대한 통제된 예시를 제공하며, ImageNet 상의 ResNet-50까지의 베이지안 신경망의 변분 사후분포(variational posterior)에서도 방향적 비교가 여전히 관찰됨을 보여준다. 코드는 https://github.com/Forsythia0604/Local-Mass-Framework 에서 확인할 수 있다.

arXiv:2606.10906v2stat.ML

Human-AI Teaming Through the Lens of Calibration

We study models for human-AI teaming through the lens of statistical calibration. We assume the team consists of an AI model and human---both of which are calibrated with respect to some partitioning of the feature space---and expose how the calibration assumptions propagate into the teaming framework. In particular, we consider frameworks that either (i) combine human and model predictions or (ii) delegate prediction responsibility to either a human or model. We show via theoretical and empirical results that existing methods for combination do not preserve the human's degree of calibration. Methods for delegation (by the very act of delegation) preserve calibration of the downstream predictors but shift the burden onto the rejector meta-model that decides who predicts. The rejector must be calibrated finely enough to locate where each member is superior, a demand that grows with the human's expertise and becomes unattainable when the human relies on information the system cannot observe.

arXiv:2606.01799v2cs.LG

Tree-Guided Identify-Then-Exploit: A Unified Framework of Best Arm Identification and Regret Minimization for Dueling Bandits

We study $N$-armed stochastic dueling bandits under the Condorcet-winner assumption with a positive global gap, where three widely adopted objectives are considered: best-arm identification (BAI), weak regret, and strong regret. We propose Tree-Guided Identify-Then-Exploit (TG-ITE), the first unified framework to tackle all these objectives to our knowledge. Without requiring transitivity assumptions, we propose a shared tree-guided identification approach to find a high-confidence incumbent within $O(N)$ comparisons. We further propose varied exploitation strategies to utilize this warm-start stage to optimize the specific objectives at hand. For a fixed global gap $\Delta$, this methodology enables our approach to (1) build the first winner-stays-style algorithm to achieve $O(N)$ weak regret; (2) realize the joint optimization of BAI and weak regret with $O(N)$ guarantees for both, eliminating the sub-optimal $O(\log N)$ factors of the existing approach; (3) attain the same $O(N\log T)$ order as specialized strong-regret approaches; and (4) achieve $O(N)$ sample complexity in BAI, matching a worst-case lower bound up to an additional $\log(1/\Delta)$ term. If a lower bound on the minimal winner gap is known, this term can be removed. Our results suggest that a single identification primitive can serve BAI and both regret objectives with linear dependence on $N$.

arXiv:2606.00419v5stat.ML

매개변수 없는 그룹 조건부 온라인 컨포멀 예측

Parameter-Free and Group Conditional Online Conformal Prediction

Quantifying the uncertainty in predictions made by machine learning models is challenging in online settings where the characteristics of the data and labels change over time. Online conformal prediction (OCP) methods tackle this problem by producing intervals for each observation and guaranteeing that a proportion of them contain their labels, a property known as marginal coverage. OCP methods, however, may systematically miscover observations belonging to groups defined by attributes like age or race, raising fairness concerns. Group-conditional OCP (G-OCP) methods address this by producing intervals that provide coverage for multiple groups. Existing approaches rely on learning rates, which control how they react to changes in the data stream, but can be hard to tune in online settings. Leveraging advancements in learning-rate-free methods, we develop a learning-rate-free G-OCP algorithm with strong group-coverage guarantees that hold under weaker assumptions than those required in prior work. Empirical results show that our method provides group coverage, produces interval lengths competitive with those of standard methods, and reacts more effectively to changes in the data than fixed-learning-rate methods.

arXiv:2605.26288v2stat.ML

Beyond Differences: Estimating and Tuning Ratio-Based Treatment Effects

When treatment effects are naturally expressed as ratios -- as in medicine, pricing, and marketing -- the ratio-based CATE $\tau(x) = \mathbb{E}[Y|W=1,X=x] / \mathbb{E}[Y|W=0,X=x]$ is the appropriate estimand. Yet the common estimators divide by an estimated baseline, impose a parametric form, or regress the outcome, where a better fit need not mean a better effect estimate. We introduce the Q-Learner, which writes $\tau(x)$ as a ratio of two odds, reducing ratio-CATE estimation for binary outcomes to two propensity classification tasks. In benchmarks on synthetic and randomized real-world data, the Q-Learner proves especially suited to industry practice. Standard hyperparameter tuning on its log-loss suffices to make it rank well, get the level of the effects right and their spread more reliably than the alternatives, whereas, for instance, the common S-Learner needs a costly, effect-specific tuning to rank well, often best, but then misses the spread of the effects. If the propensity is known, a single classifier fitted on only the converters is enough. This not only speeds up development, but is often a necessity in practice, where data are frequently collected only for converters.

arXiv:2605.15483v2stat.ME

Improving the Efficiency of Subgroup Analysis in Randomized Controlled Trials with TMLE

Subgroup analyses within randomized controlled trials are often underpowered due to limited sample sizes. We address this challenge by leveraging trial participants outside the subgroup of interest to augment estimation within the subgroup. Specifically, we study two Targeted Maximum Likelihood Estimators (TMLEs) that borrow information from non-subgroup participants within the same trial: a TMLE with pooled regression (TMLE-PR) and an Adaptive Targeted Maximum Likelihood Estimator (A-TMLE). Both estimators enable information sharing without relying on any external real-world data, thereby capitalizing on key strengths of the trial: most importantly, the protection against bias afforded by the randomized treatment, but also harmonized data collection, and consistent treatment and outcome definitions. The general strategy proposed here directly advances the priorities of key regulatory agencies, including the FDA, by improving the precision of subgroup-specific treatment effect estimates without introducing external sources of bias, thereby facilitating rigorous inference to support equitable labeling, access, and post-market evaluation. In a case study based on analysis of data from a cardiovascular outcome trial (LEADER, NCT01179048), we estimate the risk reduction of major adverse cardiac events (MACE) under liraglutide treatment among Black and Asian subgroups -- each comprising less than 10\% of the trial population -- using the proposed estimators that borrow information from the remainder of the trial. Using A-TMLE, in particular, we find estimated absolute MACE risk reductions of 1.6, 1.5, and 1.5 percentage points among Asian participants and 2.1, 2.0, and 2.1 percentage points among Black participants at 365, 540, and 730 days, respectively, with 95\% confidence intervals excluding the null at each time point.

arXiv:2605.06862v2stat.ME

Nonparametric estimation of time-varying network connections by multi-stage smoothing

We consider the problem of estimating the underlying edge probabilities of a time-varying network observed at multiple time points. The probability structure is represented by a time-varying graphon that satisfies temporal H\"older smoothness and piecewise Lipschitz conditions in the latent variables. We propose a multi-stage smoothing estimator that first applies temporal local smoothing to each edge and then performs node-domain smoothing using a data-driven neighborhood construction adapted from the method. An additional temporal smoothing step is introduced as an optional refinement when uniform accuracy over the entire time domain is required. Simulation studies demonstrate the benefits of combining temporal and node-domain smoothing under different generative models. We also apply the method to a real time-varying network dataset and show that it captures both smooth temporal evolution and structural patterns in the connectivity.

arXiv:2605.04946v3cs.LG

Training-Time Batch Normalization Reshapes Local Partition Geometry in Piecewise-Affine Networks

The effect of batch normalization (BN) on the function realized during training is less understood than its effect on optimization. We study training-time BN in continuous piecewise-affine (CPA) networks through the affine regions they induce in local $\ell_\infty$ windows around the data. Conditioned on a mini-batch, every switching hyperplane of a BN layer with learned parameters and a general CPA activation is a translate of a hyperplane through the batch centroid, at an offset given by a standardized quantity times the batch spread, independent of the raw bias. We prove an exact formula, valid under general position, for the number of regions in a window in terms of window cuts and interior intersections. For a BN layer it gives a lower bound on the local count in terms of the standardized offsets; the bound depends on the batch only through its spread and not on the biases or the batch location. Without BN the best such bound is $1$: for any weights and batch, some biases leave the window as a single region. We prove that the formula holds in deeper layers, inside parent affine regions of any rank. By the exact formula, BN increases the expected local count whenever it increases the expected numbers of window cuts and of interior intersections, which we examine empirically. Offset distributions of networks trained on synthetic data, MNIST, CIFAR-10 and TinyImageNet show higher window-cut rates under BN, and exact enumeration in two dimensions attributes most of the additional BN regions to interior intersections.

arXiv:2605.04838v2stat.ME

PAIR-CI: Calibrated Conditional Independence Testing for Causal Discovery with Incomplete Data

The standard constraint-based paradigm for causal discovery with incomplete data -- impute first, test second -- is frequently miscalibrated: any consistent conditional independence (CI) test rejects a true null with probability approaching 1 when imputation error induces spurious conditional dependence. We introduce PAIR-CI, a nonparametric CI test that restores calibration by integrating multiple imputation directly into the inferential procedure via a paired permutation design. PAIR-CI compares cross-validated models that include and exclude the candidate variable while receiving the same imputed conditioning set, so that imputation error cancels in their loss difference rather than contaminating the test statistic. A variance estimator that nests a provably consistent cross-validation estimator inside Rubin's rules yields inference that is asymptotically valid in the many-imputation limit and conservative by a derived, sign-definite mechanism -- to our knowledge, the first formal result combining these two inferential frameworks. In simulations, existing imputation-based CI tests exhibit false positive rates of 28-45% when data are missing not at random (MNAR), whereas PAIR-CI's pooled rate is at most 3.6%, exceeding the nominal 5% only in some cells with five or more conditioning variables. Integrated into the PC algorithm, PAIR-CI is stable across functional forms and missingness mechanisms, whereas every parametric pipeline we consider, including missing-value PC, has a regime in which it fails badly. On nonlinear network benchmarks it reduces structural Hamming distance by 18% (ALARM) and 11% (HAILFINDER) relative to missing-value PC and by 31-48% relative to test-wise deletion and imputation-based Fisher-$Z$ pipelines, while attaining the highest skeleton F1; when relationships are linear, a well-specified parametric pipeline remains preferable.

arXiv:2604.20219v3cs.LG

Layer-wise Geometric Approximation Rates for Deep Networks

Depth is widely viewed as a central contributor to the success of deep neural networks, yet standard approximation results typically give error bounds only for the final output of a network. Inspired by classical multiscale expansions and multigrade deep learning, we give depth a scale-dependent interpretation by constructing a single network whose intermediate readouts approximate the target at successively finer scales. For any $f\in L^p([0,1]^d)$ with $1\le p <\infty$ and any integer $N\ge2$, we construct a single network of width $3dN+2d+4$ with the same activation pattern in every hidden layer, using sine in the first two neurons and $\mathtt{ReLU}$ in the rest. For every $\ell\ge0$, the $L^p$ approximation error of the readout $\Phi_\ell$ at layer $\ell+4$ is bounded by $(2d+1)\omega_{f,p}(N^{-\ell})$, where $\omega_{f,p}$ denotes the $L^p$ translation modulus of $f$. Under the H\"older condition $\omega_{f,p}(t)\le Ct^\alpha$ with $C>0$ and $0< \alpha \le 1$, this estimate yields the geometric bound $(2d+1)C N^{-\alpha\ell}$. More generally, if $\omega_{f,p}(t)$ has a known upper bound that tends to zero as $t\downarrow0$, we obtain a stopping rule for adaptive depth selection. We can stop at the first depth where the error bound meets the desired tolerance and later append layers to meet a tighter tolerance while keeping all earlier maps and readouts fixed.

arXiv:2604.15285v2stat.ML

Structural interpretability in SVMs with truncated orthogonal polynomial kernels

Support vector machines (SVMs) can yield classifiers with similar predictive performance but different internal structures, particularly in the contributions of individual explanatory variables and their interactions. We introduce Orthogonal Representation Contribution Analysis (ORCA), a post-training methodology for SVMs constructed from tensor products of truncated Christoffel--Darboux kernels. The associated reproducing kernel Hilbert space (RKHS) admits a finite orthonormal polynomial basis in which the regularized component of the fitted decision function has an exact expansion. Squared expansion coefficients induce an additive decomposition of its squared RKHS norm. Based on this representation, normalized Orthogonal Kernel Contribution (OKC) indices quantify the fractions of the squared RKHS norm associated with total polynomial degrees, interaction orders, and input-variable contributions, both individually and through interactions of arbitrary order. ORCA summarizes these indices through joint structural maps, variable-specific displays, and comparisons of fitted classifiers. A controlled quartet of classification problems illustrates substantial structural differences under matched predictive performance. A double-spiral experiment compares five Jacobi kernels, while an application to the Vertebral Column data illustrates the method after cross-validated model selection. ORCA provides a quantitative characterization of the internal structure of fitted SVMs, revealing differences that may remain undetected by standard predictive performance metrics.

arXiv:2604.13410v3stat.ME

Estimating Continuous Treatment Effects with Two-Stage Kernel Ridge Regression

We study the problem of estimating the effect function for a continuous treatment, which maps each treatment value to a population-averaged outcome. A central challenge in this setting is confounding: treatment assignment often depends on covariates, creating selection bias that makes direct regression of the response on treatment unreliable. To address this issue, we propose a two-stage kernel ridge regression method. In the first stage, we learn a model for the response as a function of both treatment and covariates; in the second stage, we use this model to construct pseudo-outcomes that correct for distribution shift, and then fit a second model to estimate the treatment effect. Although the response varies with both treatment and covariates, the induced effect function obtained by averaging over covariates is typically much simpler, and our estimator adapts to this structure. Our optimal learning bounds are achieved without estimating the conditional treatment density, thereby bypassing a major bottleneck in existing methods. Furthermore, we introduce a fully data-driven model selection procedure that achieves provable adaptivity to both the unknown degree of overlap and the spectral decay of the underlying kernel.

arXiv:2603.27049v3stat.ML

인센티브 붕괴 역설 극복하기

Overcoming the Incentive Collapse Paradox

AI 보조 작업 위임은 점차 일반화되고 있으나, 이러한 시스템에서 인간의 노력은 비용이 많이 들고 통상 관측 불가능합니다. Bastani와 Cachon (2025); Sambasivan 등 (2021)의 최근 연구는 정확도 기반 지불 방식이 인센티브 붕괴(incentive collapse) 문제를 겪는다는 것을 보여줍니다. 즉, AI 정확도가 향상됨에 따라 긍정적인 인간의 노력을 유지하려면 무한한 지불이 필요합니다. 우리는 출력 정확도가 관측되지 않은 노력에 의존하는 전략적 인간 에이전트(agent)가 포함된 예산 제약이 있는 주인-에이전트(principal-agent) 프레임워크 내에서 이 현상을 연구합니다. 첫 번째 기여는 인센티브 붕괴가 단순한 선형 지불의 한계에 불과한 것이 아니라, 관측된 작업 정확도에만 기반한 모든 지불 규칙에 대해 발생한다는 것을 보여주는 일반적인 불가능성 결과입니다. 이러한 장벽을 극복하기 위해, 우리는 AI 정확도와 무관하게 유한한 비용으로 인간의 노력 수준을 엄격하게 양의 값이고 제어 가능한 수준으로 강제하는 감시자 감사(sentinel-auditing) 지불 메커니즘을 제안합니다. 이 인센티브 강건한 기반을 바탕으로, 우리는 단일 예산 하에서 최종 통계적 손실(loss)을 최소화하기 위해 (i) 감사율과 (ii) 다양한 난이도의 작업에 대한 능동적 샘플링 및 예산 할당을 공동으로 최적화하는 인센티브 인지 능동적 통계 추론(inference) 프레임워크를 개발합니다. 실험 결과, 표준 능동 학습 및 감사 전용 기준선에 비해 개선된 비용-오차 트레이드오프를 입증합니다.

arXiv:2603.17896v2stat.ML

A Noise Sensitivity Exponent Controls Large Statistical-to-Computational Gaps in Single- and Multi-Index Models

Understanding when learning is statistically possible yet computationally hard is a central challenge in high-dimensional statistics. In this work, we investigate this question in the context of single- and multi-index models, classes of functions widely studied as benchmarks to probe the ability of machine learning methods to discover features in high-dimensional data. Our main contribution is to show that a Noise Sensitivity Exponent (NSE) - a simple quantity determined by the activation function - governs the existence and magnitude of statistical-to-computational gaps within a broad regime of these models. We first establish that, in single-index models with large additive noise, the onset of a computational bottleneck is fully characterized by the NSE. We then demonstrate that the same exponent controls a statistical-computational gap in the specialization transition of large separable multi-index models, where individual components become learnable. Finally, in hierarchical multi-index models, we show that the NSE governs the optimal computational rate in which different directions are sequentially learned. Taken together, our results identify the NSE as a unifying property linking noise robustness, computational hardness, and feature specialization in high-dimensional learning.

arXiv:2603.07955v5math.GT

RL unknotter, hard unknot과 unknotting number

RL unknotter, hard unknots and unknotting number

We train a reinforcement learning agent to simplify hard unknot diagrams and pair it with a learned crossing selector to search for short unknotting sequences. Three examples found by the agent lead to an infinite family of prime two-bridge knots of unknotting number three whose connected sums with the figure-eight knot also have unknotting number three, and fifteen new unknotting upper bounds were found on the set of prime knots up to 13 crossings.

arXiv:2603.03035v2stat.ML

Generalized Bayes for Causal Inference

Uncertainty quantification is central to many applications of causal machine learning, yet principled Bayesian inference for causal effects remains challenging. Standard Bayesian approaches typically require specifying a probabilistic model for the data-generating process, including high-dimensional nuisance components such as propensity scores and outcome regressions. Standard posteriors are thus vulnerable to strong modeling choices, including complex prior elicitation. In this paper, we propose a generalized Bayesian framework for causal inference. Our framework avoids explicit likelihood modeling; instead, we place priors directly on the causal estimands and update these using an identification-driven loss function, which yields generalized posteriors for causal effects. As a result, our framework turns existing loss-based causal estimators into estimators with full uncertainty quantification. Our framework is flexible and applicable to a broad range of causal estimands (e.g., ATE, CATE). Further, our framework can be applied on top of state-of-the-art causal machine learning pipelines (e.g., Neyman-orthogonal meta-learners). For Neyman-orthogonal losses, we show that the generalized posteriors converge to their oracle counterparts and remain robust to first-stage nuisance estimation error. With calibration, we thus obtain valid frequentist uncertainty even when nuisance estimators converge at slower-than-parametric rates. Empirically, we demonstrate that our proposed framework offers causal effect estimation with calibrated uncertainty across several causal inference settings. To the best of our knowledge, this is the first flexible framework for constructing generalized Bayesian posteriors for causal machine learning.

arXiv:2603.02275v3cs.LG

A Comparative Study of UMAP and Other Dimensionality Reduction Methods

Uniform Manifold Approximation and Projection (UMAP) is a widely used manifold learning technique for dimensionality reduction. This paper studies UMAP, supervised UMAP, and several competing dimensionality reduction methods, including Principal Component Analysis (PCA), Kernel PCA, Sliced Inverse Regression (SIR), Kernel SIR, and t-distributed Stochastic Neighbor Embedding, through a comprehensive comparative analysis. Although UMAP has attracted substantial attention for preserving local and global structures, its supervised extensions, particularly for regression settings, remain rather underexplored. We provide a systematic evaluation of supervised UMAP for both regression and classification using simulated and real datasets, with performance assessed via predictive accuracy on low-dimensional embeddings. Our results show that supervised UMAP performs well for classification but exhibits limitations in effectively incorporating response information for regression, highlighting an important direction for future development.

arXiv:2602.22505v2cs.LG

Sharp Convergence Rates for Masked Diffusion Models

Discrete diffusion models have achieved strong empirical performance in text and other symbolic domains, with masked (absorbing-rate) variants emerging as competitive alternatives to autoregressive models. Among existing samplers, the Euler method remains the standard choice in many applications, and more recently, the First-Hitting Sampler (FHS) has shown considerable promise for masked diffusion models. Despite their practical success, the theoretical understanding of these samplers remains incomplete. Many existing analyses are conducted in Kullback-Leibler (KL) divergence, which often yields loose parameter dependencies and requires strong assumptions on score estimation. Moreover, these guarantees do not cover recently developed high-performance sampler of FHS. In this work, we first develop a direct total-variation (TV)-based analysis for the Euler method that overcomes these limitations. Our results relax assumptions on score estimation, improve parameter dependencies, and establish convergence guarantees without requiring any surrogate initialization. Also for this setting, we provide the first convergence lower bound for the Euler sampler, establishing tightness with respect to all of the data dimension $d$, the vocabulary size $S$, and the target accuracy $\varepsilon$, up to logarithmic factors. Finally, we analyze the FHS sampler and show that it incurs no sampling error beyond that induced by score estimation, which we show to be tight with a matching lower error bound. Our analysis introduces two techniques that may be of independent interest: a comparison with a reference unmasking schedule for the Euler method and a pathwise argument exploiting the decoupling structure of FHS.