Searcharxiv⌕ Search

arXiv subjects

Fan Li

Publications and source records attributed to Fan Li.

At least 19 recordsLinked to original sources
arXiv2026

Correlation Matters! Streamlining the Sample Size Procedure with Hierarchical Time-to-event Endpoints

Composite endpoints are widely used in cardiovascular clinical trials to improve statistical efficiency while preserving clinical relevance. The win ratio (WR) measure and more general frameworks of win statistics have emerged as popular alternatives to traditional time-to-first-event analyses. Although analytic sample size formulas for WR have been developed, they rely on design parameters that are often not straightforward to specify. Consequently, sample size determination in clinical trials with WR for the primary analysis is most often based on simulations, which, however, can be computationally intensive. To address this challenge, we derive refined formulas to calculate the proportions of wins, losses, and ties for multiple prioritized time-to-event endpoints to assist with study planning. These formulas rely on familiar design inputs and become directly applicable when integrated with existing sample size methods. We conduct a comprehensive assessment of how correlation among endpoints affects sample size requirements across varying design features, and demonstrate the role of correlations through case studies to generate further insights.

stat.ME↗
arXiv2026

WeSearch: A Unified Model-Centric Retrieval Framework for WeChat Search

We present WeSearch, a unified, model-centric retrieval framework for large-scale, continuously evolving content search. Prevailing methods assign documents fixed hierarchical Semantic IDs, implicitly assuming each document follows a single coherent semantic path-an assumption that often breaks down for multi-topic documents. WeSearch instead retrieves query-relevant clusters and then ranks atomic documents within the resulting query-specific candidate space, decoupling candidate pruning from fine-grained relevance modeling. Combined with relevance-aware training, compact document representations, and incremental index updates, this design enables weekly-or daily when needed-model refreshes. Deployed as the primary semantic retrieval path in WeChat Search alongside a lexical syntax tree, WeSearch shows consistent one-week A/B gains on corpora from hundreds of millions to billions of documents, and a one-month full-traffic A/B test confirms stable improvements of +0.68% CTR and +2.27% per-user article clicks, together with a 41% reduction in retrieval resource cost and a 19 ms reduction in graphic-text retrieval latency. WeSearch is now fully deployed in WeChat Search, serving 100% of production traffic.

cs.IR↗
arXiv2026

Two-stage least squares with treatment-covariate interactions for treatment effect heterogeneity

Treatment effect heterogeneity with respect to covariates is common in instrumental variable (IV) analyses. An intuitive approach, which we call the interacted two-stage least squares (2sls), is to regress the outcome on the treatment, covariates, and treatment-covariate interactions, using the IV, covariates, and IV-covariate interactions as instruments. We clarify the causal interpretation of the interacted 2sls under the local average treatment effect (LATE) framework when the IV is valid conditional on the covariates. Our main findings are twofold. First, we show that the coefficients on the treatment-covariate interactions from the interacted 2sls identify treatment effect heterogeneity with respect to covariates among compliers for any outcome-generating process if and only if the product of the IV propensity score and the covariates is linear in the covariates, a requirement we call the linear IV-covariate interactions condition. Second, when the covariate vector has dimension K and includes a constant term, we show that the linear IV-covariate interactions condition requires the IV propensity score to take at most K distinct values. The condition holds when (a) the covariates take exactly K distinct values, or (b) the IV is randomly assigned independently of the covariates, but is restrictive beyond these cases. These results underscore the difficulty of interpreting regression coefficients from specifications with treatment-covariate interactions when the covariates are not saturated and the IV is not unconditionally randomized, absent correct specification of the outcome model. We then connect these results to the literature on LATE estimation, and show that under the linear IV-covariate interactions condition, the interacted 2sls with covariates demeaned by their complier means recovers the LATE as a single coefficient.

stat.ME↗
arXiv2026

PlanViz: Evaluating Planning-Oriented Image Generation and Editing for Computer-Use Tasks

Unified multimodal models (UMMs) have shown impressive capabilities in complex natural image generation. However, their potential in computer-use planning tasks, which are closely related to our lives, remains rarely explored. Image generation and editing in computer-use tasks require capabilities like fine-grained perception and procedural understanding, and it is still unknown whether UMMs can finish these tasks well. Therefore, we propose PlanViz, a new benchmark designed to evaluate image generation and editing for computer-use tasks. We focus on sub-tasks which frequently involve in daily life and require planning steps. Specifically, three representative sub-tasks are designed: route planning, work diagramming, and web\&UI displaying. Data quality is ensured by including large human efforts on question, reference and criteria annotation. Additionally, a task-adaptive score, PlanScore, is proposed for comprehensive and exact evaluation. PlanScore helps understanding the correctness, visual quality and efficiency of generated images. Through experiments, we highlight key limitations and opportunities for future research on this topic. Codes and supplementary materials are in https://github.com/lijunxian111/PlanViz.

cs.CV↗
arXiv2026

Can discrete-time analyses be trusted for stepped wedge trials with continuous recruitment?

In stepped wedge cluster randomized trials (SW-CRTs), the intervention is rolled out to clusters sequentially over multiple periods. It is common practice to analyze data from SW-CRTs using linear mixed models that treat time as discrete. However, a recent systematic review found that 95.1\% of cross-sectional SW-CRTs recruit individuals continuously over time. Despite the prevalence of continuous recruitment, there has been limited guidance on how to draw model-robust inference for such designs. In this article, we conduct simulation studies to examine the performance of discrete-time linear mixed models in the case of continuous recruitment designs with a continuous outcome. Specifically, we generate outcomes under a continuous-time exponential decay correlation structure with either discrete or continuous fixed period effects. We consider constant and exposure-time-dependent intervention effects, with and without random intervention effects across clusters. We then analyze the simulated data using linear mixed models under three discrete-time working correlation structures: simple exchangeable, nested exchangeable, and discrete-time exponential decay, with robust sandwich variance estimation. Our results demonstrate that discrete-time analysis often yields negligible bias and that the robust variance estimator with the Mancl and DeRouen correction consistently achieves nominal coverage and type I error rate. However, when recruitment patterns vary systematically between control and intervention periods, discrete-time analyses give slightly biased intervention effect estimates. Finally, we illustrate these findings by reanalyzing a completed SW-CRT.

stat.ME↗
arXiv2026

A Bayesian machine learning approach to direct and spillover effects in cluster-randomized trials with noncompliance

We develop a Bayesian machine learning approach for estimating direct and spillover effects in cluster-randomized trials with treatment noncompliance. To flexibly model outcomes and treatment receipt while accounting for within-cluster dependence, we combine Bayesian additive regression trees with cluster-level random effects. We augment the outcome regression with a peer-treatment clever covariate, defined as the predicted mean proportion of treated peers from a preliminary receipt model, to mitigate regularization-induced distortion of the effect decomposition. Within a Bayesian g-computation framework, we estimate cluster-average intent-to-treat effects and cluster-average local average treatment effects and examine sensitivity to the cross-world restrictions underlying their decomposition. Simulation studies show reduced bias under model misspecification, while the correctly specified parametric models show expected frequentist properties. We find that the clever covariate improves the decomposition of the total effect into direct and spillover components under moderate or strong spillover, with similar gains in smaller clusters. Its benefits vary with spillover strength, and the $95\%$ credible intervals remain conservative. An application to the Primary School Deworming Project in Kenya suggests reduced infection risk, with a larger estimated contribution from individual treatment receipt than from peer receipt. Overall, our approach provides a flexible Bayesian estimation of direct and spillover effect decomposition in the presence of treatment noncompliance.

stat.ME↗
arXiv2026

Where to weight? Estimating population causal effects with weighted double score matching in complex surveys

Double score matching is an emerging method for causal inference that combines propensity and prognostic scores to achieve dimension reduction and model robustness. However, when the observational data are drawn from a complex survey, the existing literature does not provide sufficient guidance on the proper use of survey weights in matching strategies for population-level causal inference. We develop weighted double score matching (WDSM) that provides a unified framework for incorporating survey weights throughout the score estimation, matching, and outcome analysis under both retrospective (informative) and prospective (non-informative) sampling designs. We establish asymptotic normality of the resultant effect estimates under both known and estimated scores and develop a linearization-based bootstrap procedure for variance estimation that preserves the survey-weighted matching structure. Simulation studies show that WDSM substantially reduces bias relative to unweighted matching when either the propensity score model or the prognostic score models are correctly specified. Under correct specification, WDSM also improves efficiency and confidence interval coverage relative to survey-weighted propensity score matching, with particularly pronounced gains under poor overlap. Applications to early marijuana use and mental health in the National Survey on Drug Use and Health and to sleep duration and glycemic control in the National Health and Nutrition Examination Survey illustrate the practical value of WDSM for population-level inference. We conclude with a discussion of our findings and provide practical guidance on incorporating survey weights into matching-based analyses of observational data.

stat.ME↗
arXiv2026

Dynamic Regret and Tightened Constraint Violation for Communication-Efficient Distributed Online Bilevel Optimization

This paper investigates the distributed online bilevel optimization problem with time-varying inequality constraints. In this setting, a multi-agent network is confronted with a hierarchical decision-making task characterized by coupling between the inner and outer optimization levels. We first propose a communication-efficient single-loop distributed online primal--dual algorithm based on communication compression, which tracks both the inner-level solution and the Hessian-inverse-vector product using only one update per-round. We show that this algorithm achieves a dynamic network regret bound of $\mathcal{O}(T^{\max\{θ_1,1-θ_1\}}+T^{θ_1 } P_T+H_{1,T}+E_{1,T})$, where~$θ_1 \in (0,1)$, $P_T$ is the outer-level path variation, $H_{1,T}$ is the inner-level path variation, and $E_{1,T}$ is the gradient path variation alongside a network cumulative constraint violation bound of $\mathcal{O}(T^{1-θ_1/2})$, which is further reduced to $\mathcal{O}(T^{1-θ_1})$ under Slater's condition. Furthermore, to eliminate the dependence of regret on the inner-level and gradient path variation, we propose a novel adaptive multi-loop distributed online algorithm to control the inner-level approximation accuracy at each round. By dynamically adjusting the number of inner-level updates based on a gradient-norm tolerance condition, this algorithm yields an improved dynamic network regret bound of $\mathcal{O}(T^{\max\{θ_1,1-θ_1\}}+T^{θ_1 } P_T)$ which depends solely on the outer-level path variation while maintaining the same network cumulative constraint violation bound. Finally, numerical simulations verify the effectiveness of the proposed algorithms.

math.OC↗
arXiv2026

FocusOPD: Structure-Guided On-Policy Distillation for Diffusion Models

Multi-teacher on-policy distillation enables rectified-flow models with diffusion-transformer architectures to combine task-specific capabilities within a single student. However, matching model outputs does not guarantee that their internal representations are aligned, and it remains unclear where and when additional feature supervision is most useful. We analyze the teacher-student feature mismatch across Transformer blocks and denoising timesteps, and find that the mismatch is concentrated in deeper Transformer blocks during early denoising and persists after OPD distillation, indicating ineffective distillation. Motivated by this observation, we introduce FocusOPD, which augments the base OPD objective with token-wise directional feature alignment restricted to deep layers and early denoising steps. Across two rectified-flow backbones and multiple OPD objectives, FocusOPD improves capability transfer in compositional generation, text rendering, and preference-based evaluation. On the primary benchmark, it improves over DiffusionOPD from 0.953 to 0.967 on GenEval, from 23.797 to 23.944 on PickScore, and from 6.052 to 6.199 on LAION-Aesthetic. Ablations show that selective deep and early supervision is more effective and efficient than alignment over the full network and trajectory.

cs.CV↗
arXiv2026

When Pairwise Preferences Cycle: A Reliability-Anchored Preference Walk for Video GRPO

Pairwise preference feedback has become a widely adopted supervision signal for reinforcement learning in video generation models, as pairwise comparisons provide a more reliable alternative than pointwise scalar ratings for evaluating the quality of generated videos. Specifically, we define a preference graph for each GRPO group under a pairwise evaluator, where nodes represent sampled rollouts and edges encode pairwise preferences. Auditing 6,976 on-policy comparison graphs, we find that cyclic preferences are prevalent and persist throughout training. Naive pairwise reward aggregation based on average win rates discards the underlying comparison structures and treats each rollout independently, which may introduce unreliable advantage estimates for certain rollouts and consequently affect reinforcement learning optimization. Therefore, we propose the \emph{Reliability-Anchored Preference Walk} (\ours{}), a graph-aware advantage aggregation method that strengthens the global preference signal from win rates with reliable relational evidence from cyclic preference graphs through a reliability-guided preference walk. Under matched evaluator, comparison protocol, group size, prompt schedule, and query budget, \ours{} requires no additional evaluator queries beyond the observed comparisons within each rollout group and improves over Pairwise-GRPO across controlled comparisons, transfer settings, and blind human evaluations.

cs.CV↗
arXiv2026

Calibrated Bayesian inference for overall and subgroup-specific causal effects in stepped-wedge cluster-randomized trials

Stepped-wedge cluster-randomized trials (SW-CRTs) randomly assign clusters to sequences that determine when each cluster switches from control to intervention. Conventional Bayesian analyses of SW-CRTs report treatment effect coefficients from a working outcome model, but under model misspecification or informative cluster-period sizes, such coefficients can lack a clear causal interpretation. In this work, we develop Bayesian methods targeting weighted average treatment effect estimands in SW-CRTs, including novel subgroup-specific causal estimands defined by baseline characteristics. We compare model-based g-computation with model-robust standardization, implementing both with linear mixed models and mixed-effects Bayesian additive regression trees as working outcome models, and propose a calibrated variance estimator that augments posterior parameter uncertainty with cluster-level resampling to better capture sampling variability. In simulations with outcome model misspecification and informative cluster-period sizes, model-robust standardization with calibrated inference exhibits negligible bias and near-nominal frequentist coverage for overall and subgroup-specific causal effects, whereas g-computation and uncalibrated intervals become insufficient. We illustrate the proposed methods with a reanalysis of the Trauma Survivors Outcomes and Support trial.

stat.ME↗
arXiv2026

Totem: Object-Centric 3D Memory for World-Consistent Video Generation

Generating coherent videos grounded in a unified underlying world remains a major challenge, particularly under large viewpoint changes, where existing methods often struggle to maintain stable representations of individual objects. To this end, we introduce Totem, an object-centric 3D memory framework for world-consistent video generation. Totem first represents each selected object as a persistent 3D memory aligned to a shared world frame, assigning it a dedicated state. Based on these memories, we design a dual-stream 3D conditioning architecture that combines object memories with scene geometry for target-view synthesis. Moreover, these object memories constrain the model's attention with 3D structural information, thereby preserving object surfaces and relative spatial layouts. Extensive experiments on world-generation and object-centric benchmarks demonstrate that Totem generates videos depicting a more object-consistent 3D world under challenging camera trajectories, while supporting versatile applications.

cs.CV↗
arXiv2026

Design-based theory for covariate adjustment in generalized pairwise comparisons

Generalized pairwise comparisons summarize treatment effects in randomized experiments with ordinal, multivariate, or prioritized outcomes. We study optimal covariate adjustment under complete randomization, with potential outcomes and covariates fixed and treatment assignment as the sole source of randomness. We derive the exact randomization variance of the unadjusted estimator and characterize linear calibration adjustments using the between-arm difference in covariate means. The asymptotic variance is a strictly convex quadratic in the adjustment coefficient, yielding a closed-form optimum. The resulting estimator is asymptotically no less efficient than the unadjusted estimator, with an explicit precision gain that clarifies which covariates improve precision. However, natural pair-level and single-arm least squares adjustments generally use weights that differ from the allocation-dependent optimum and can substantially reduce efficiency. We also propose a within-arm variance estimator for asymptotically conservative inference, whose conservatism arises from unidentifiable joint variation in a unit's two potential scores, and show that residualizing each arm on its own projection coefficient saves the discrepancy that governs the efficiency loss of conventional adjustments. In the special case where the comparison function is the outcome difference, our proposed covariate-adjusted estimator is first-order equivalent to the fully interacted regression adjustment of Lin (Ann. Appl. Stat., 2013, 7, 295-318).

stat.ME↗
arXiv2026

DSR-Net: Learning Dynamic Skeletal Relations for 3D Human Pose Estimation

3D human pose estimation requires accurate modeling of motion-dependent skeletal structures and inter-joint dependencies. Existing methods typically rely on fixed skeletal topologies in Graph Convolutional Networks (GCNs) or implicitly capture joint relations through attention mechanisms, limiting their adaptability to diverse motion patterns. To address this issue,we proposeDSR-Net, a dynamic structural representation learning framework for 3D human pose estimation. Specifically, a Dynamic Skeletal Relation (DSR) module is introduced to learn adaptive inter-joint dependencies beyond predefined skeletal connections. Based on the learned relations, a hierarchical relation-aware fusion strategy is developed to facilitate effective interaction between structural and temporal representations. In addition, a lightweight Residual Skeletal Recalibration Module (RSRM) is designed to enhance the stability of structure-aware features with minimal computational overhead. Extensive experiments on Human3.6M and MPI-INF-3DHP demonstrate that DSR-Net achieves MPJPEs of 37.9 mm and 14.9 mm, respectively, consistently improving over strong baselines. The results validate the effectiveness of dynamic skeletal relation modeling for robust 3D human pose estimation.

cs.CV↗
arXiv2026

On the permutation equivariance principle for causal estimands

In many causal inference problems, multiple action variables share a common causal role yet lack a natural ordering. \revblue{We consider $K\geq2$ action variables, each evaluated under treatment or control conditions,} and formalize permutation equivariance, the principle that permuting the variables permutes the corresponding estimands in a trackable manner, hence preserving their scientific meaning. We characterize this principle algebraically and present a complete class of weighted permutation equivariant estimands capturing main effects and interactions of all orders. We discuss the interpretation and choice of weights and characterize residual-free estimands, whose inclusion--exclusion sum recovers the endpoint contrast between the all-treated and all-control configurations. \revblue{Applying our general framework to network interference yields a new hierarchy of direct, indirect, and overall effects, whose first-order aggregates recover the average effects of \citet{hu2022average}. We also identify the overall effects as mixed derivatives of expected welfare under independent Bernoulli assignment.} We illustrate the framework through the contexts of factorial studies, causal mediation, and network interference.

stat.ME↗
arXiv2026

Sparse Attention Is Matrix Approximation, Not Choosing from a Bag of Values

Large Language Models (LLMs) achieve strong performance across many domains, but their efficiency is limited by the quadratic cost of attention with respect to prompt length. Sparse attention reduces this cost by retaining only a small fraction of query-key interactions to approximate the full attention matrix. However, existing methods are trapped in a mathematically wrong view: they simply keep large scalar entries or high-mass regions of the attention matrix. This treats the attention matrix as a bag of values, ignoring that it is used as a structured matrix whose entries jointly determine the attention output through multiplication with value vectors. We argue that this is the core conceptual issue: sparse attention should be formulated as matrix approximation, not as blindly choosing the largest values from a bag of entries. Based on this view, we propose Matrix Approximation Sparse Attention (MASA). MASA replaces raw attention-mass ranking with a closed-form score that measures how much each sparse unit reduces matrix-product approximation error. As a theory-grounded plug-in correction, MASA can be added to existing sparse attention frameworks without changing their sparse kernels or budgets. Extensive experiments across multiple sparse attention methods, benchmarks, and LLM backbones show consistent accuracy gains, supporting both MASA and the matrix-approximation view of sparse attention.

cs.CL↗
arXiv2026

Do Higher-Order Models Win for Higher-Order Reasons? Rethinking Performance Gains in Hypergraph Learning

Higher-order models (e.g., hypergraph neural networks) often outperform lower-order baselines on hypergraph learning benchmarks, and their advantages are commonly attributed to their ability to exploit higher-order information. However, better performance alone does not establish this explanation. We therefore ask: Do higher-order models win for higher-order reasons? To investigate this question, we introduce a controlled performance-attribution framework that perturbs higher-order information while preserving the lower-order, i.e., pairwise, information. Across 25 commonly used hypergraph learning benchmarks spanning three tasks, we frequently observe an intriguing pattern: higher-order models originally outperform lower-order baselines, yet retain most of their advantage after perturbation. This suggests that much of the observed advantage remains achievable without the higher-order information. We then investigate potential lower-order explanations for these remaining gaps. We find that simple additions to a lower-order baseline, e.g., richer pairwise weighting, more steps of pairwise feature propagation, and normalization, reduce the remaining performance gaps, supporting lower-order explanations for part of the observed advantage. Our analysis calls for the hypergraph learning community to rethink performance attribution by distinguishing performance gains from their explanations, adopt stronger lower-order baselines, and use suitable benchmarks that better test the value of higher-order information.

cs.LG↗
arXiv2026

General theory of persistent microwave-optical quantum resources in hybrid-system dynamics

We develop a general theoretical framework for characterizing persistent quantum resources between microwave and optical modes in the dynamics of chain-type hybrid quantum systems with intermediate modes. The effective Hamiltonian for microwave-optical (MO) squeezing is formulated via strong nearest-neighbor interactions in the microwave-intermediate-optical chain, from which rigorous solutions for the dynamics of MO Gaussian entanglement and quantum steering are obtained analytically. Notably, MO quantum resource can survive and approach a finite asymptotic value even in the unsteady regime, and can surpass the steady-state upper bound on the quantum resource. Furthermore, the asymptotic values of MO entanglement as well as one-way and two-way quantum steering are readily controllable by tuning the effective coupling strength. The validity of our theory is demonstrated by applying it to the typical hybrid models of electro-optomechanical and cavity optomagnomechanical systems, and the extension to the nonlinear regime with non-Gaussian MO quantum resources is further studied.

quant-ph↗