Searcharxiv⌕ Search

arXiv subjects

Shu Yang

Publications and source records attributed to Shu Yang.

At least 55 records · Page 3Linked to original sources
arXiv2026

Spatial causal inference in the presence of preferential sampling to study the impacts of marine protected areas

Marine Protected Areas (MPAs) have been established globally to conserve marine resources. Given their maintenance costs and impact on commercial fishing, it is critical to evaluate their effectiveness to support future conservation. In this paper, we use data collected from the Australian coast to estimate the effect of MPAs on biodiversity. Environmental studies such as these are often observational, and processes of interest exhibit spatial dependence, which presents challenges in estimating the causal effects. Spatial data can also be subject to preferential sampling, where the sampling locations are related to the policy and the response variable, further complicating inference and prediction. To address these challenges, we propose a spatial causal inference method that simultaneously accounts for unmeasured spatial confounders in both the sampling process and the treatment allocation. We prove the identifiability of key parameters in the model and the consistency of the posterior distributions of those parameters. We show via simulation studies that the causal effect of interest can be reliably estimated under the proposed model. The proposed method is applied to assess the effect of MPAs on fish biomass. We find evidence of preferential sampling and that properly accounting for this source of bias impacts the estimate of the causal effect.

stat.ME↗
arXiv2026

A Semantic-Sampling Framework for Evaluating Calibration in Open-Ended Question Answering

Calibration measures whether a model's predicted confidence aligns with its empirical accuracy, and is central to the reliable deployment of large language models (LLMs) in high-stakes domains such as medicine and law. While much recent work focuses on improving LLM calibration, the equally important question of how to evaluate it in realistic settings remains underdeveloped. Open-ended question answering (QA), the most common deployment setting for modern LLMs, is where existing evaluation methods fall short: logit-based metrics need restricted output formats and internal probabilities; verbalized confidence is self-reported and often overconfident; and sampling-based methods rely on task-specific extraction rules without a clear finite-sample target. We introduce Sem-ECE (Semantic-Sampling Expected Calibration Error), a calibration evaluation framework for open-ended QA that samples answers from the model, groups them into semantic classes, and uses the resulting frequencies as confidence. We study two estimators within this framework: Sem$_1$-ECE, the same-sample self-consistency score, and Sem$_2$-ECE, a held-out variant that separates answer selection from confidence evaluation. We prove both are asymptotically unbiased, and further show that they agree on easy questions but diverge on hard ones with Sem$_2$ achieving strictly smaller calibration error, so their gap also serves as a diagnostic for question difficulty. Experiments on three open-ended QA benchmarks across five leading commercial LLMs match our theoretical predictions and show that Sem-ECE outperforms verbalized confidence and existing sampling-based methods, while complementing logit-based evaluation when internal probabilities are unavailable.

cs.CL↗
arXiv2026

History-Aware Conformal Prediction Sets for Censored Time-to-Event Outcomes

Existing conformal prediction methods for time-to-event outcomes leverage only baseline covariates, producing prediction intervals that are insufficiently informative to facilitate decision making. We propose History-Aware Prediction Sets (HAPS), a conformal framework that constructs prediction sets for individual event times using covariate histories observed up to a decision time, targeting coverage among individuals who have survived to this time. HAPS handles right censoring adjusted for time-varying confounders via inverse probability of censoring weighting. When the censoring weights are consistently estimated, it achieves PAAC (probably asymptotically approximately correct) coverage among survivors. We further propose two doubly robust extensions of HAPS to weaken reliance on consistent estimation of the censoring distribution. In simulations, HAPS and its extensions reduce median prediction interval length by up to 75\% relative to baseline comparators while maintaining close to nominal coverage. On two public benchmark data sets, HAPS reduces the median interval length by up to 60\% for predictions at year 5, compared to the baseline comparators.

stat.ME↗
arXiv2026

Externally Controlled Trials: A Review of Design and Borrowing Through a Causal Lens

Externally controlled trials (ECTs) are increasingly used when randomized controls are infeasible, unethical, or insufficient, including applications in rare diseases, oncology, pediatrics, and post-approval effectiveness research. Although methodological work has expanded rapidly across causal inference, Bayesian dynamic borrowing, and hybrid trial designs, the literature remains fragmented. We adopt a six-step scientific roadmap to organize modern ECT methodology in two primary settings: (i) single-arm trials that evaluate efficacy through comparison with external controls, and (ii) hybrid controlled trials that augment the internal control arm with external controls drawn from real-world data or historical studies. The roadmap clarifies causal estimands, identifiability assumptions, and how statistical parameters arise from identification, and shows how modeling and borrowing strategies trade off efficiency and robustness, especially under covariate shift and outcome drift. Within this framework, we synthesize and evaluate recent Bayesian and frequentist developments, compare their strengths, limitations, operating characteristics, and available software, and emphasize the role of sensitivity analysis. By re-framing ECT methodology through a causal lens, this work establishes a coherent foundation for integrating external data into regulatory and clinical decision-making and highlights core challenges and opportunities for future research.

stat.ME↗
arXiv2026

Adaptive Influence-Based Borrowing Framework for Improving Treatment Effect Estimation in RCTs Using External Controls

Randomized controlled trials (RCTs) often suffer from limited sample sizes due to high costs and lengthy recruitment periods, compromising precision in treatment effect estimation. External real-world control data offer a valuable opportunity for augmentation, but naïve integration may introduce bias without careful compatibility assessment. This paper presents a practical tutorial on the adaptive influence-based borrowing framework~\citep{Yang-etal2026}, which addresses this challenge through a principled, individual-level borrowing strategy. The core intuition is straightforward: rather than indiscriminately pooling all external controls (ECs), the framework first asks how much each external patient would perturb the outcome model fitted using RCT controls. External patients whose inclusion barely changes this model are deemed comparable and prioritized for borrowing, whereas those who substantially shift it are flagged as potentially incompatible. This individual-level compatibility metric, based on the influence score, is then used to construct a sequence of nested candidate subsets of ECs, from which the optimal subset is selected by minimizing the mean squared error of the treatment effect estimator, balancing the competing risks of bias from over-borrowing and imprecision from under-borrowing. When systematic differences between ECs and RCT controls are substantial, an optional outcome calibration step can align the two groups before influence-based selection proceeds. We provide a clear, step-by-step workflow with emphasis on methodological intuition, practical considerations, and visualization, thereby offering a principled, transparent, and practical method for leveraging ECs when RCTs alone are underpowered. Implementation is supported by an accompanying \texttt{R} package InfluenceBorrowing.

stat.AP↗
arXiv2026

Flattery in Motion: Benchmarking and Analyzing Sycophancy in Video-LLMs

As video large language models (Video-LLMs) become increasingly integrated into real-world applications that demand grounded multimodal reasoning, ensuring their factual consistency and reliability is of critical importance. However, sycophancy, the tendency of these models to align with user input even when it contradicts the visual evidence, undermines their trustworthiness in such contexts. Current sycophancy research has largely overlooked its specific manifestations in the videolanguage domain, resulting in a notable absence of systematic benchmarks and targeted evaluations to understand how Video-LLMs respond under misleading user input. To fill this gap, we propose VISE(Video-LLM Sycophancy Benchmarking and Evaluation), the first benchmark designed to evaluate sycophantic behavior in state-of-the-art Video-LLMs across diverse question formats, prompt biases, and visual reasoning tasks. Specifically, VISEpioneeringly brings linguistic perspectives on sycophancy into the video domain, enabling fine-grained analysis across multiple sycophancy types and interaction patterns. Furthermore, we propose two potential training-free mitigation strategies revealing potential paths for reducing sycophantic bias: (i) enhancing visual grounding through interpretable key-frame selection and (ii) steering model behavior away from sycophancy via targeted, inference-time intervention on its internal neural representations. Our code is available at https://anonymous.4open.science/r/VideoSycophancy-567F.

cs.CL↗
arXiv2026

A Survey on the Safety and Security Threats of Computer-Using Agents: JARVIS or Ultron?

Recently, AI-driven interactions with computing devices have advanced from basic prototype tools to sophisticated, LLM-based systems that emulate human-like operations in graphical user interfaces. We are now witnessing the emergence of \emph{Computer-Using Agents} (CUAs), capable of autonomously performing tasks such as navigating desktop applications, web pages, and mobile apps. However, as these agents grow in capability, they also introduce novel safety and security risks. Vulnerabilities in LLM-driven reasoning, with the added complexity of integrating multiple software components and multimodal inputs, further complicate the security landscape. In this paper, we present a systematization of knowledge on the safety and security threats of CUAs. We conduct a comprehensive literature review and distill our findings along four research objectives: \textit{\textbf{(i)}} define the CUA that suits safety analysis; \textit{\textbf{(ii)} } categorize current safety threats among CUAs; \textit{\textbf{(iii)}} propose a comprehensive taxonomy of existing defensive strategies; \textit{\textbf{(iv)}} summarize prevailing benchmarks, datasets, and evaluation metrics used to assess the safety and performance of CUAs. Building on these insights, our work provides future researchers with a structured foundation for exploring unexplored vulnerabilities and offers practitioners actionable guidance in designing and deploying secure Computer-Using Agents.

cs.CL↗
arXiv2026

A Robust Framework for Two-Sample Mendelian Randomization under Population Heterogeneity

Mendelian randomization is a powerful tool for causal inference in observational studies. The two-sample summary-data design, which estimates genetic associations with exposures and outcomes in separate cohorts, is the most widely used Mendelian randomization approach in large-scale genomic studies. However, this approach relies on a strong assumption of population homogeneity across the two samples. In practice, available samples often differ in ancestry, demographics, socioeconomic factors, covariate adjustment, and measurement protocols. Violations of the homogeneity assumption can bias causal effect estimates and undermine the credibility of Mendelian randomization findings. We introduce a robust, model-free Mendelian randomization framework that directly addresses population heterogeneity in the two-sample summary-data setting. Our method avoids parametric assumptions about population differences and is designed to address real-world challenges, including measurement error, weak instruments, and pleiotropy. We show that the proposed estimator is consistent and asymptotically normal under heterogeneous designs, and may offer efficiency gains over the classic estimator even in homogeneous settings. Through numerical simulations and a real data analysis for estimating the causal effect of body mass index on high-density lipoprotein cholesterol across ancestrally diverse populations, we demonstrate the practical utility, stability, and robustness of our approach.

stat.ME↗
arXiv2026

Multi-User Large Language Model Agents

Large language models (LLMs) and LLM-based agents are increasingly deployed as assistants in planning and decision making, yet most existing systems are implicitly optimized for a single-principal interaction paradigm, in which the model is designed to satisfy the objectives of one dominant user whose instructions are treated as the sole source of authority and utility. However, as they are integrated into team workflows and organizational tools, they are increasingly required to serve multiple users simultaneously, each with distinct roles, preferences, and authority levels, leading to multi-user, multi-principal settings with unavoidable conflicts, information asymmetry, and privacy constraints. In this work, we present the first systematic study of multi-user LLM agents. We begin by formalizing multi-user interaction with LLM agents as a multi-principal decision problem, where a single agent must account for multiple users with potentially conflicting interests and associated challenges. We then introduce a unified multi-user interaction protocol and design three targeted stress-testing scenarios to evaluate current LLMs' capabilities in instruction following, privacy preservation, and coordination. Our results reveal systematic gaps: frontier LLMs frequently fail to maintain stable prioritization under conflicting user objectives, exhibit increasing privacy violations over multi-turn interactions, and suffer from efficiency bottlenecks when coordination requires iterative information gathering.

cs.CL↗
arXiv2026

End-pinching and inertial-capillary reopening in viscoplastic ligaments at low Ohnesorge number

Capillary retraction of liquid ligaments is well understood for Newtonian fluids, whereas viscoplastic effects remain comparatively unexplored. Here, we consider Herschel-Bulkley fluids, which incorporate both yield stress and shear-rate-dependent viscosity, thereby introducing a spatially varying effective viscosity that is absent in simpler yield-stress models (e.g., Bingham models). We focus on the low-viscosity regime, where droplet detachment in Newtonian fluids is controlled by the end-pinching mechanism. Using fully resolved axisymmetric simulations, we show that viscoplasticity and shear-rate-dependent rheology reorganize the routes by which a retracting ligament may pinch off, escape break-up or stay motionless due to large yield stress. We identify two distinct routes by which a retracting Herschel-Bulkley ligament can escape end-pinching. In the shear-thickening regime, increased local viscosity during neck thinning leads to larger vorticity detachment from the curved neck, which opposes the capillary singularity. In the strongly shear-thinning regime, reopening is governed by curvature-induced pressure gradients. We show that this latter mechanism persists in the Newtonian limit of vanishing viscosity, yielding a purely inertial-capillary pathway for reopening. While previous Newtonian studies report end-pinching down to Ohnesorge number $Oh_K \approx 10^{-4}$, suggesting break-up as the asymptotic low-viscosity outcome (Anthony et al. 2019), our results demonstrate that a purely inertial-capillary reopening mechanism can arise as $Oh_K \to 0$, indicating that end-pinching is not the route in the inviscid limit.

physics.flu-dyn↗
arXiv2026

Visual Self-Fulfilling Alignment: Shaping Safety-Oriented Personas via Threat-Related Images

Multimodal large language models (MLLMs) face safety misalignment, where visual inputs enable harmful outputs. To address this, existing methods require explicit safety labels or contrastive data; yet, threat-related concepts are concrete and visually depictable, while safety concepts, like helpfulness, are abstract and lack visual referents. Inspired by the Self-Fulfilling mechanism underlying emergent misalignment, we propose Visual Self-Fulfilling Alignment (VSFA). VSFA fine-tunes vision-language models (VLMs) on neutral VQA tasks constructed around threat-related images, without any safety labels. Through repeated exposure to threat-related visual content, models internalize the implicit semantics of vigilance and caution, shaping safety-oriented personas. Experiments across multiple VLMs and safety benchmarks demonstrate that VSFA reduces the attack success rate, improves response quality, and mitigates over-refusal while preserving general capabilities. Our work extends the self-fulfilling mechanism from text to visual modalities, offering a label-free approach to VLMs alignment.

cs.CV↗
arXiv2026

Improving Treatment Effect Estimation in Trials through Adaptive Borrowing of External Controls

Randomized controlled trials (RCTs) often suffer from limited inferential efficiency in estimating treatment effects due to their small sample sizes. In recent years, incorporating external controls (ECs) has gained increasing attention as an effective way to augment small RCTs and thereby enhance estimation efficiency. However, ECs are not always comparable to RCTs, and direct borrowing without careful evaluation can introduce substantial bias and, paradoxically, undermine the accuracy of treatment effect estimation. In this paper, we propose a novel adaptive influence-based sample borrowing framework to improve average treatment effect (ATE) estimation in RCTs. The framework quantifies the ``comparability'' of each sample in ECs using influence functions and identifies the optimal subset of ECs that minimizes the mean squared error of the ATE estimator. The proposed framework is assumption-lean regarding the distribution of ECs and is robust to outliers, making it broadly applicable across diverse settings. Moreover, we develop an outcome calibration method to improve the data utilization efficiency of ECs, further strengthening the adaptive influence-based sample-borrowing framework. We demonstrate the effectiveness of the proposed method using both simulated and real-world datasets.

stat.ME↗
arXiv2026

STARS: Skill-Triggered Audit for Request-Conditioned Invocation Safety in Agent Systems

Autonomous language-model agents increasingly rely on installable skills and tools to complete user tasks. Static skill auditing can expose capability surface before deployment, but it cannot determine whether a particular invocation is unsafe under the current user request and runtime context. We therefore study skill invocation auditing as a continuous-risk estimation problem: given a user request, candidate skill, and runtime context, predict a score that supports ranking and triage before a hard intervention is applied. We introduce STARS, which combines a static capability prior, a request-conditioned invocation risk model, and a calibrated risk-fusion policy. To evaluate this setting, we construct SIA-Bench, a benchmark of 3,000 invocation records with group-safe splits, lineage metadata, runtime context, canonical action labels, and derived continuous-risk targets. On a held-out split of indirect prompt injection attacks, calibrated fusion reaches 0.439 high-risk AUPRC, improving over 0.405 for the contextual scorer and 0.380 for the strongest static baseline, while the contextual scorer remains better calibrated with 0.289 expected calibration error. On the locked in-distribution test split, gains are smaller and static priors remain useful. The resulting claim is therefore narrower: request-conditioned auditing is most valuable as an invocation-time risk-scoring and triage layer rather than as a replacement for static screening. Code is available at https://github.com/123zgj123/STARS.

cs.AI↗
arXiv2026

Hierarchical Alignment: Enforcing Hierarchical Instruction-Following in LLMs through Logical Consistency

Large language models increasingly operate under multiple instructions from heterogeneous sources with different authority levels, including system policies, user requests, tool outputs, and retrieved context. While prior work on instruction hierarchy highlights the importance of respecting instruction priorities, it mainly focuses on adversarial attacks and overlooks the benign but common instruction conflicts that arise in real-world applications. In such settings, models must not only avoid security violations but also preserve task utility and behavioral consistency when instructions partially or implicitly conflict. We propose Neuro-Symbolic Hierarchical Alignment (NSHA) for hierarchical instruction-following by explicitly modeling and enforcing instruction priorities. At inference time, we introduce solver-guided reasoning that formulates instruction resolution as a constraint satisfaction problem, enabling the model to derive a maximally consistent set of applicable instructions under hierarchical constraints. At training time, NSHA distills solver-based decisions into model parameters using automatically constructed supervision. We evaluate our approach on rule following, task execution, tool use, and safety, covering both single-turn and multi-turn interactions, and show that NSHA significantly improves performance under such conflicts while maintaining competitive utility in reference settings.

cs.CL↗
arXiv2026

Sample-split REGression SREG: A robust estimator for high-dimensional survey data

Model-assisted regression estimation is fundamental in survey sampling for incorporating auxiliary information. However, when the auxiliary dimension grows with the sample size, the standard Generalized regression (GREG) estimator can exhibit non-negligible bias under informative sampling, even when the working model is correctly specified. This failure stems from the double use of sampled outcomes simultaneously for fitting the regression and for forming the residual correction. We propose a sample-split REGression (SREG) estimator based on K-fold cross-fitting that eliminates this bias by pairing each unit's residual with an out-of-fold prediction. The resulting estimator is first-order equivalent to the oracle difference estimator under a weak prediction-norm consistency requirement, without requiring root-n consistent estimation of regression coefficients. We establish asymptotic normality and prove consistency of a variance estimator based on cross-fitted residuals. The key conditional fluctuation assumption is verified for simple random, stratified, and rejective sampling. Simulations demonstrate that SREG effectively removes high-dimensional bias while maintaining competitive efficiency.

stat.ME↗
arXiv2026

Observation of Floquet erratic non-Hermitian skin effect in photonic mesh lattice

In ordered, translationally invariant non-Hermitian systems, the skin effect is understood as a boundary phenomenon: nonreciprocal hopping drives an extensive accumulation of eigenstates towards the edges, whereas the periodic-boundary spectrum remains Bloch extended. Here we experimentally reveal the opposite limit -- a disorder-enabled, boundary-independent, and intrinsically bulk form of skin localization -- the recently predicted erratic non-Hermitian skin effect (ENHSE), realized in a driven photonic platform. Using a time-multiplexed photonic mesh lattice with programmable gain, loss, and phase modulation, we engineer spatially fluctuating imaginary gauge fields and realize a Floquet non-Hermitian lattice whose global reciprocity can be tuned independently of strong local nonreciprocity. We observe a disorder-driven non-Hermitian topological transition between two oppositely directed disordered skin phases through a critical point of global reciprocity. At this transition, boundary skin accumulation disappears, yet the wave dynamics self-organizes into bulk-localized patterns without any interface, providing direct evidence of ENHSE. The measured localization profiles agree with simulations and exhibit the defining feature that distinct eigenstates share a common bulk-localized envelope determined by the disordered imaginary gauge fields. By further introducing controllable on-site disorder, we reveal the competition between ENHSE and Anderson localization, and show how increasing scattering progressively suppresses erratic skin dynamics. Our results help establish ENHSE as a unique disorder-induced non-Hermitian phenomenon and open a route to engineering localization, transport, and topology beyond conventional Bloch and boundary-based paradigms.

physics.optics↗
arXiv2026

Concept-Based Dictionary Learning for Inference-Time Safety in Vision Language Action Models

Vision Language Action (VLA) models close the perception action loop by translating multimodal instructions into executable behaviors, but this very capability magnifies safety risks: jailbreaks that merely yield toxic text in LLMs can trigger unsafe physical actions in embodied systems. Existing defenses alignment, filtering, or prompt hardening intervene too late or at the wrong modality, leaving fused representations exploitable. We introduce a concept based dictionary learning framework for inference time safety control. By learning sparse, interpretable dictionaries from hidden activations, our method identifies harmful concept directions and attenuates risky components when the estimated risk exceeds a threshold. Experiments on Libero-Harm, BadRobot, RoboPair, and IS-Bench show that our approach achieves state-of-the-art defense performance, cutting attack success rates by over 70\% while maintaining task success. Crucially, the framework is plug-in and model-agnostic, requiring no retraining and integrating seamlessly with diverse VLAs. To our knowledge, this is the first inference time concept based safety method for embodied systems, advancing both interpretability and safe deployment of VLA models.

cs.RO↗
arXiv2026

Can AI Truly Represent Your Voice in Deliberations? A Comprehensive Study of Large-Scale Opinion Aggregation with LLMs

Large-scale public deliberations generate thousands of free-form contributions that must be synthesized into representative and neutral summaries for policy use. While LLMs have been shown as a promising tool to generate summaries for large-scale deliberations, they also risk underrepresenting minority perspectives and exhibiting bias with respect to the input order, raising fairness concerns in high-stakes contexts. Studying and fixing these issues requires a comprehensive evaluation at a large scale, yet current practice often relies on LLMs as judges, which show weak alignment with human judgments. To address this, we present DeliberationBank, a large-scale human-grounded dataset with (1) opinion data spanning ten deliberation questions created by 3,000 participants and (2) summary judgment data annotated by 4,500 participants across four dimensions (representativeness, informativeness, neutrality, policy approval). Using these datasets, we train DeliberationJudge, a fine-tuned DeBERTa model that can rate deliberation summaries from individual perspectives. DeliberationJudge is more efficient and more aligned with human judgements compared to a wide range of LLM judges. With DeliberationJudge, we evaluate 18 LLMs and reveal persistent weaknesses in deliberation summarization, especially underrepresentation of minority positions. Our framework provides a scalable and reliable way to evaluate deliberation summarization, helping ensure AI systems are more representative and equitable for policymaking.

cs.CL↗