arXiv subjects

Wenxin Zhang

Publications and source records attributed to Wenxin Zhang.

At least 37 records · Page 2Linked to original sources

arXiv2026

An Online Meta-Level Adaptive Design Framework with Targeted Learning Inference: Applications to Evaluating and Utilizing Surrogate Outcomes in Adaptive Designs

Adaptive designs are increasingly used in clinical trials and online experiments to improve participant outcomes by dynamically updating treatment allocation as data accumulate. In practice, experimenters often consider multiple candidate designs, each with distinct trade-offs. However, typically only one design is implemented at a time, leaving benefits and costs of alternative designs unobserved and unquantified. To address this, we propose a novel meta-level adaptive design framework that enables real-time, data-driven evaluation and selection among candidate adaptive designs. Specifically, we define a new class of causal estimands to evaluate adaptive designs and propose Targeted Maximum Likelihood Estimators for these estimands. These estimators are asymptotically normal while accommodating dependence in adaptive-design data without parametric assumptions, enabling online selection among candidate designs. We further apply this framework to a motivating example where multiple surrogates of a long-term outcome are considered for updating randomization probabilities in adaptive experiments. Unlike existing surrogate evaluation methods, our approach comprehensively quantifies surrogates' utility to accelerate detection of heterogeneous treatment effects, expedite updates to treatment randomization, and improve participant outcomes, facilitating dynamic selection among surrogate-guided designs. Overall, our framework provides a unified approach for evaluating opportunities and costs of various adaptive designs and guiding real-time decision-making in adaptive experiments.

stat.ME↗

arXiv2026

GraphiContact: Pose-aware Human-Scene Robust Contact Perception for Interactive Systems

Monocular vertex-level human-scene contact prediction is a fundamental capability for interactive systems such as assistive monitoring, embodied AI, and rehabilitation analysis. In this work, we study this task jointly with single-image 3D human mesh reconstruction, using reconstructed body geometry as a scaffold for contact reasoning. Existing approaches either focus on contact prediction without sufficiently exploiting explicit 3D human priors, or emphasize pose/mesh reconstruction without directly optimizing robust vertex-level contact inference under occlusion and perceptual noise. To address this gap, we propose GraphiContact, a pose-aware framework that transfers complementary human priors from two pretrained Transformer encoders and predicts per-vertex human-scene contact on the reconstructed mesh. To improve robustness in real-world scenarios, we further introduce a Single-Image Multi-Infer Uncertainty (SIMU) training strategy with token-level adaptive routing, which simulates occlusion and noisy observations during training while preserving efficient single-branch inference at test time. Experiments on five benchmark datasets show that GraphiContact achieves consistent gains on both contact prediction and 3D human reconstruction. Our code, based on the GraphiContact method, provides comprehensive 3D human reconstruction and interaction analysis, and will be publicly available at https://github.com/Aveiro-Lin/GraphiContact.

cs.CV↗

arXiv2026

Exploring the Design of GenAI-Based Systems to Support Socially Shared Metacognition

Socially shared metacognition (SSM) refers to the collective monitoring and regulation of joint cognitive processes in collaborative problem-solving, and is essential for effective knowledge work and learning. Generative AI (GenAI)-based systems offer new opportunities to support SSM, but emerging evidence suggests that poorly designed systems can encourage over-reliance on AI-generated explicit instruction and erode groups' capacity to develop autonomous regulatory processes. Group awareness tools (GATs) address this challenge through established design principles that make social and cognitive awareness information visible, highlight differences between group members to create cognitive conflict, and trigger autonomous elaboration and discussion, thereby implicitly guiding autonomous SSM emergence. This paper explores the design of GenAI-augmented GATs to support autonomous SSM in collaborative work and learning through an initial literature search, presenting preliminary design principles for discussion.

cs.HC↗

arXiv2026

Design Guidance Towards Addressing Over-Reliance on AI in Sensemaking

Sensemaking in collaborative work and learning is increasingly supported by GenAI systems, however, emerging evidence suggests that poorly designed GenAI systems tend to provide explicit instruction that groups passively follow, fostering over-reliance and eroding autonomous sensemaking. Group awareness tools (GATs) address this challenge through implicit guidance: rather than instructing groups on what to do, GATs externalize observable collaboration data through visualizations that reveal differences between group members to create cognitive conflict, which triggers autonomous elaboration and discussion, thereby implicitly guiding autonomous sensemaking emergence. Drawing on an initial literature search of existing GAT systems, this paper explores the design of GenAI-augmented GATs to support autonomous sensemaking in collaborative work and learning, presenting preliminary design principles for discussion.

cs.HC↗

arXiv2026

SPARK: Jailbreaking T2V Models by Synergistically Prompting Auditory and Recontextualized Knowledge

Jailbreak attacks can circumvent model safety guardrails and reveal critical blind spots. Prior attacks on text-to-video (T2V) models typically add adversarial perturbations to obviously unsafe prompts, which are often easy to detect and defend. In contrast, we show that benign-looking prompts containing rich, implicit cues can induce T2V models to generate semantically unsafe videos that both violate policy and preserve the original (blocked) intent. To realize this, we propose SPARK, a jailbreak framework that leverages T2V models cross-modal associative patterns via a modular prompt design. Specifically, our prompts combine three components: neutral scene anchors, which provide the surface-level scene description extracted from the blocked intent to maintain plausibility; latent auditory triggers, textual descriptions of innocuous-sounding audio events (e.g., creaking, muffled noises) that exploit learned audio-visual co-occurrence priors to bias the model toward particular unsafe visual concepts; and stylistic modulators, cinematic directives (e.g., camera framing, atmosphere) that amplify and stabilize the latent trigger's effect. We formalize attack generation as a constrained optimization over the above modular prompt space and solve it with a guided search procedure that balances stealth and effectiveness. Extensive experiments over 7 T2V models demonstrate the efficacy of our attack, achieving a +23% improvement in average attack success rate in commercial models.

cs.CV↗

arXiv2026

DeepSVU: Towards In-depth Security-oriented Video Understanding via Unified Physical-world Regularized MoE

In the literature, prior research on Security-oriented Video Understanding (SVU) has predominantly focused on detecting and localize the threats (e.g., shootings, robberies) in videos, while largely lacking the effective capability to generate and evaluate the threat causes. Motivated by these gaps, this paper introduces a new chat paradigm SVU task, i.e., In-depth Security-oriented Video Understanding (DeepSVU), which aims to not only identify and locate the threats but also attribute and evaluate the causes threatening segments. Furthermore, this paper reveals two key challenges in the proposed task: 1) how to effectively model the coarse-to-fine physical-world information (e.g., human behavior, object interactions and background context) to boost the DeepSVU task; and 2) how to adaptively trade off these factors. To tackle these challenges, this paper proposes a new Unified Physical-world Regularized MoE (UPRM) approach. Specifically, UPRM incorporates two key components: the Unified Physical-world Enhanced MoE (UPE) Block and the Physical-world Trade-off Regularizer (PTR), to address the above two challenges, respectively. Extensive experiments conduct on our DeepSVU instructions datasets (i.e., UCF-C instructions and CUVA instructions) demonstrate that UPRM outperforms several advanced Video-LLMs as well as non-VLM approaches. Such information.These justify the importance of the coarse-to-fine physical-world information in the DeepSVU task and demonstrate the effectiveness of our UPRM in capturing such information.

cs.CV↗

arXiv2026

WaveSP-Net: Learnable Wavelet-Domain Sparse Prompt Tuning for Speech Deepfake Detection

Modern front-end design for speech deepfake detection relies on full fine-tuning of large pre-trained models like XLSR. However, this approach is not parameter-efficient and may lead to suboptimal generalization to realistic, in-the-wild data types. To address these limitations, we introduce a new family of parameter-efficient front-ends that fuse prompt-tuning with classical signal processing transforms. These include FourierPT-XLSR, which uses the Fourier Transform, and two variants based on the Wavelet Transform: WSPT-XLSR and Partial-WSPT-XLSR. We further propose WaveSP-Net, a novel architecture combining a Partial-WSPT-XLSR front-end and a bidirectional Mamba-based back-end. This design injects multi-resolution features into the prompt embeddings, which enhances the localization of subtle synthetic artifacts without altering the frozen XLSR parameters. Experimental results demonstrate that WaveSP-Net outperforms several state-of-the-art models on two new and challenging benchmarks, Deepfake-Eval-2024 and SpoofCeleb, with low trainable parameters and notable performance gains. The code and models are available at https://github.com/xxuan-acoustics/WaveSP-Net.

eess.AS↗

arXiv2026

Mask-GCG: Are All Tokens in Adversarial Suffixes Necessary for Jailbreak Attacks?

Jailbreak attacks on Large Language Models (LLMs) have demonstrated various successful methods whereby attackers manipulate models into generating harmful responses that they are designed to avoid. Among these, Greedy Coordinate Gradient (GCG) has emerged as a general and effective approach that optimizes the tokens in a suffix to generate jailbreakable prompts. While several improved variants of GCG have been proposed, they all rely on fixed-length suffixes. However, the potential redundancy within these suffixes remains unexplored. In this work, we propose Mask-GCG, a plug-and-play method that employs learnable token masking to identify impactful tokens within the suffix. Our approach increases the update probability for tokens at high-impact positions while pruning those at low-impact positions. This pruning not only reduces redundancy but also decreases the size of the gradient space, thereby lowering computational overhead and shortening the time required to achieve successful attacks compared to GCG. We evaluate Mask-GCG by applying it to the original GCG and several improved variants. Experimental results show that most tokens in the suffix contribute significantly to attack success, and pruning a minority of low-impact tokens does not affect the loss values or compromise the attack success rate (ASR), thereby revealing token redundancy in LLM prompts. Our findings provide insights for developing efficient and interpretable LLMs from the perspective of jailbreak attacks.

cs.CL↗

arXiv2025

Data reuse enables cost-efficient randomized trials of medical AI models

Randomized controlled trials (RCTs) are indispensable for establishing the clinical value of medical artificial-intelligence (AI) tools, yet their high cost and long timelines hinder timely validation as new models emerge rapidly. Here, we propose BRIDGE, a data-reuse RCT design for AI-based risk models. AI risk models support a broad range of interventions, including screening, treatment selection, and clinical alerts. BRIDGE trials recycle participant-level data from completed trials of AI models when legacy and updated models make concordant predictions, thereby reducing the enrollment requirement for subsequent trials. We provide a practical checklist for investigators to assess whether reusing data from previous trials allows for valid causal inference and preserves type I error. Using real-world datasets across breast cancer, cardiovascular disease, and sepsis, we demonstrate concordance between successive AI models, with up to 64.8% overlap in top 5% high-risk cohorts. We then simulate a series of breast cancer screening studies, where our design reduced required enrollment by 46.6%--saving over US$2.8 million--while maintaining 80% power. By transforming trials into adaptive, modular studies, our proposed design makes Level I evidence generation feasible for every model iteration, thereby accelerating cost-effective translation of AI into routine care.

cs.LG↗

arXiv2025

Constructing Confidence Intervals for Infinite-Dimensional Functional Parameters by Highly Adaptive Lasso

Estimating the conditional mean function is a central task in statistical learning. In this paper, we consider estimation and inference for a nonparametric class of real-valued cadlag functions with bounded sectional variation (Gill et al., 1995), using the Highly Adaptive Lasso (HAL) (van der Laan, 2015; Benkeser and van der Laan, 2016; van der Laan, 2023), a flexible empirical risk minimizer over linear combinations of tensor products of zero- or higher-order spline basis functions under an L1 norm constraint. Building on recent theoretical advances in asymptotic normality and uniform convergence rates for higher-order spline HAL estimators, this work focuses on constructing robust confidence intervals for HAL-based estimators of conditional means. First, we propose a targeted HAL with a debiasing step to remove the regularization bias of the targeted conditional mean and also consider a relaxed HAL estimator to reduce such bias within the working model. Second, we propose both global and local undersmoothing strategies to adaptively enlarge the working model and further reduce bias relative to variance. Third, we combine these estimation strategies with delta-method-based variance estimators to construct confidence intervals for the conditional mean. Through extensive simulation studies, we evaluate different combinations of our estimation procedures, model selection strategies, and confidence-interval constructions. The results show that our proposed approaches substantially reduce bias relative to variance and yield confidence intervals with coverage rates close to nominal levels across different scenarios. Finally, we demonstrate the general applicability of our framework by estimating conditional average treatment effect (CATE) functions, highlighting how HAL-based inference methods extend to other infinite-dimensional, non-pathwise-differentiable parameters.

stat.ME↗

arXiv2025

Tail-Optimized Caching for LLM Inference

Prompt caching is critical for reducing latency and cost in LLM inference: OpenAI and Anthropic report up to 50-90% cost savings through prompt reuse. Despite its widespread success, little is known about what constitutes an optimal prompt caching policy, particularly when optimizing tail latency, a metric of central importance to practitioners. The widely used Least Recently Used (LRU) policy can perform arbitrarily poor on this metric, as it is oblivious to the heterogeneity of conversation lengths. To address this gap, we propose Tail-Optimized LRU, a simple two-line modification that reallocates KV cache capacity to prioritize high-latency conversations by evicting cache entries that are unlikely to affect future turns. Though the implementation is simple, we prove its optimality under a natural stochastic model of conversation dynamics, providing the first theoretical justification for LRU in this setting, a result that may be of independent interest to the caching community. Experimentally, on real conversation data WildChat, Tail-Optimized LRU achieves up to 27.5% reduction in P90 tail Time to First Token latency and 23.9% in P95 tail latency compared to LRU, along with up to 38.9% decrease in SLO violations of 200ms. We believe this provides a practical and theoretically grounded option for practitioners seeking to optimize tail latency in real-world LLM deployments.

eess.SY↗

arXiv2025

HAL-Based Plug-in Estimation with Pointwise Asymptotic Normality of the Causal Dose-Response Curve

Estimating and obtaining reliable inference for the marginally adjusted causal dose-response curve for continuous treatments without relying on parametric assumptions is a well-known statistical challenge. Parametric models risk introducing significant bias through model misspecification, compromising the accurate representation of the underlying data and dose-response relationship. On the other hand, nonparametric models face difficulties as the dose-response curve is not pathwise differentiable, preventing consistent estimation at standard rates. The Highly Adaptive Lasso (HAL) maximum likelihood estimator offers a promising approach to this issue. In this paper, we introduce a HAL-based plug-in estimator for the causal dose-response curve, bridge theoretical development and empirical application, and assess its empirical performance against other estimators. This work emphasizes not just theoretical proofs, but also demonstrates their application through comprehensive simulations, thereby filling an essential gap between theory and practice. Our comprehensive simulations demonstrate that the HAL-based estimator achieves pointwise asymptotic normality with valid inference and consistently outperforms existing approaches for estimating the causal dose-response curve.

stat.ME↗

arXiv2025

Efficient Statistical Estimation for Sequential Adaptive Experiments with Implications for Adaptive Designs

Adaptive experimental designs have gained popularity in clinical trials and online experiments. Unlike traditional, fixed experimental designs, adaptive designs can dynamically adjust treatment randomization probabilities and other design features in response to data accumulated sequentially during the experiment. These adaptations are useful to achieve diverse objectives, including reducing uncertainty in the estimation of causal estimands or increasing participants' chances of receiving better treatments during the experiment. At the end of the experiment, it is often desirable to answer causal questions from the observed data. However, the adaptive nature of such experiments and the resulting dependence among observations pose significant challenges to providing valid statistical inference and efficient estimation of causal estimands. Building upon the Targeted Maximum Likelihood Estimator (TMLE) framework tailored for adaptive designs (van der Laan, 2008), we introduce a new Adaptive-Design-Likelihood-based TMLE (ADL-TMLE) to estimate a wide class of causal estimands from adaptive experiment data, including the average treatment effect as our primary example. We establish asymptotic normality and semiparametric efficiency of ADL-TMLE under relaxed positivity and design stabilization assumptions for adaptive experiments. Motivated by these results, we further propose a novel adaptive design aimed at minimizing the variance of the estimator based on data generated under that design. Simulations show that ADL-TMLE demonstrates superior variance-reduction performance across different types of adaptive experiments, and that the proposed adaptive design attains lower variance than the standard efficiency-oriented adaptive design. Finally, we generalize our framework to broader settings, including those with longitudinal structures.

stat.ME↗

arXiv2025

Fake-Mamba: Real-Time Speech Deepfake Detection Using Bidirectional Mamba as Self-Attention's Alternative

Advances in speech synthesis intensify security threats, motivating real-time deepfake detection research. We investigate whether bidirectional Mamba can serve as a competitive alternative to Self-Attention in detecting synthetic speech. Our solution, Fake-Mamba, integrates an XLSR front-end with bidirectional Mamba to capture both local and global artifacts. Our core innovation introduces three efficient encoders: TransBiMamba, ConBiMamba, and PN-BiMamba. Leveraging XLSR's rich linguistic representations, PN-BiMamba can effectively capture the subtle cues of synthetic speech. Evaluated on ASVspoof 21 LA, 21 DF, and In-The-Wild benchmarks, Fake-Mamba achieves 0.97%, 1.74%, and 5.85% EER, respectively, representing substantial relative gains over SOTA models XLSR-Conformer and XLSR-Mamba. The framework maintains real-time inference across utterance lengths, demonstrating strong generalization and practical viability. The code is available at https://github.com/xuanxixi/Fake-Mamba.

eess.AS↗

arXiv2025

Butter: Frequency Consistency and Hierarchical Fusion for Autonomous Driving Object Detection

Hierarchical feature representations play a pivotal role in computer vision, particularly in object detection for autonomous driving. Multi-level semantic understanding is crucial for accurately identifying pedestrians, vehicles, and traffic signs in dynamic environments. However, existing architectures, such as YOLO and DETR, struggle to maintain feature consistency across different scales while balancing detection precision and computational efficiency. To address these challenges, we propose Butter, a novel object detection framework designed to enhance hierarchical feature representations for improving detection robustness. Specifically, Butter introduces two key innovations: Frequency-Adaptive Feature Consistency Enhancement (FAFCE) Component, which refines multi-scale feature consistency by leveraging adaptive frequency filtering to enhance structural and boundary precision, and Progressive Hierarchical Feature Fusion Network (PHFFNet) Module, which progressively integrates multi-level features to mitigate semantic gaps and strengthen hierarchical feature learning. Through extensive experiments on BDD100K, KITTI, and Cityscapes, Butter demonstrates superior feature representation capabilities, leading to notable improvements in detection accuracy while reducing model complexity. By focusing on hierarchical feature refinement and integration, Butter provides an advanced approach to object detection that achieves a balance between accuracy, deployability, and computational efficiency in real-time autonomous driving scenarios. Our model and implementation are publicly available at https://github.com/Aveiro-Lin/Butter, facilitating further research and validation within the autonomous driving community.

cs.CV↗

arXiv2025

Super Ensemble Learning Using the Highly-Adaptive-Lasso

We introduce the Meta Highly-Adaptive-Lasso Minimum Loss Estimator (M-HAL-MLE), a novel ensemble approach for estimating functional parameters of realistically modeled data distribution from independent and identically distributed observations. Given $J$ initial estimators, candidate ensembles are generated by finite-sectional-variation cadlag functions. Using $V$-fold cross-validation, the M-HAL-MLE selects the optimal cadlag ensemble minimizing the cross-validated empirical risk, with the sectional variation bound as a tuning parameter. The final estimator, M-HAL super-learner, is obtained by averaging ensemble compositions across folds. In contrast, the oracle ensemble and oracle estimator are defined by minimizing the population excess risk relative to the true function. We establish following theoretical properties: 1) the M-HAL super-learner converges to the oracle estimator at rate $n^{-2/3}$ in excess risk, up to log-n factors; 2) by appropriate undersmoothing, target features of the M-HAL super-learner are asymptotically linear for corresponding target features of the oracle estimator; 3) the excess risk between the oracle estimator and true function, along with the difference between their target features, is generally second-order. Simulations validate the theoretical results, demonstrating effectiveness in high-dimensional settings. We further illustrate the method in a real-data application involving mediation analysis of functional MRI from human pain studies.

stat.ME↗

arXiv2025

Dynamic Pricing for Reusable Resources: The Power of Two Prices

Motivated by real-world applications such as rental and cloud computing services, we investigate pricing for reusable resources. We consider a system where a single resource with a fixed number of identical copies serves customers with heterogeneous willingness-to-pay (WTP), and the usage duration distribution is general. Optimal dynamic policies are computationally intractable when usage durations are not memoryless, so existing literature has focused on static pricing, which incurs a steady-state performance loss of ${O}(\sqrt{c})$ compared to optimality when supply and demand scale with $c$. We propose a class of dynamic "stock-dependent" policies that 1) are computationally tractable and 2) can attain a steady-state performance loss of $o(\sqrt{c})$. We give parametric bounds based on the local shape of the reward function at the optimal fluid admission probability and show that the performance loss of stock-dependent policies can be as low as ${O}((\log{c})^2)$. We characterize the tight performance loss for stock-dependent policies and show that they can in fact be achieved by a simple two-price policy that sets a higher price when the stock is below some threshold and a lower price otherwise. We extend our results to settings with multiple resources and multiple customer classes. Finally, we demonstrate this "minimally dynamic" class of two-price policies performs well numerically, even in non-asymptotic settings, suggesting that a little dynamicity can go a long way.

math.OC↗

arXiv2025

FreCT: Frequency-augmented Convolutional Transformer for Robust Time Series Anomaly Detection

Time series anomaly detection is critical for system monitoring and risk identification, across various domains, such as finance and healthcare. However, for most reconstruction-based approaches, detecting anomalies remains a challenge due to the complexity of sequential patterns in time series data. On the one hand, reconstruction-based techniques are susceptible to computational deviation stemming from anomalies, which can lead to impure representations of normal sequence patterns. On the other hand, they often focus on the time-domain dependencies of time series, while ignoring the alignment of frequency information beyond the time domain. To address these challenges, we propose a novel Frequency-augmented Convolutional Transformer (FreCT). FreCT utilizes patch operations to generate contrastive views and employs an improved Transformer architecture integrated with a convolution module to capture long-term dependencies while preserving local topology information. The introduced frequency analysis based on Fourier transformation could enhance the model's ability to capture crucial characteristics beyond the time domain. To protect the training quality from anomalies and improve the robustness, FreCT deploys stop-gradient Kullback-Leibler (KL) divergence and absolute error to optimize consistency information in both time and frequency domains. Extensive experiments on four public datasets demonstrate that FreCT outperforms existing methods in identifying anomalies.

cs.LG↗