arXiv subjects

Xinkai Wang

Publications and source records attributed to Xinkai Wang.

11 recordsLinked to original sources

arXiv2026

Track4Action: Distilling World-Centric 3D Tracker into Vision-Language-Action Policies

Action labels tell a vision-language-action (VLA) policy which robot commands to imitate, but not how those commands change the 3D world. The aligned demonstration clip contains this missing supervision because its $K$ frame transitions record the geometry, motion, visibility, and camera change produced during the corresponding $K$ actions. We introduce Track4Action, a framework that distills this realized transition from a frozen world-centric 3D tracker into a current-observation VLA policy. During training, Track4World encodes the clip $V_{t:t+K}$ into a pooled tracker feature. Learnable track queries infer this feature from current VLA hidden states, match it in a shared space, and condition a flow-matching action head through a feature-wise gate. The tracker feature only defines the alignment target, so neither the clip nor the tracker is used at deployment. Track4Action reaches 82.3% on zero-shot LIBERO-Plus, improving the alignment-free variant by 7.6 points and LaMP by 3.0 points. It obtains 80.44% and 81.48% on the clean and randomized RoboTwin 2.0 splits, and 67.5% average success across four physical bimanual tasks, 25.0 points above the alignment-free variant. The gains across simulation and physical tasks support action-aligned 3D tracker features as privileged supervision for tracker-free VLA deployment. Our project page is available at https://wing0night.github.io/track4action-project-page.

cs.RO↗

arXiv2026

When Is a Conformal Guarantee Fair? Auditing Silent Subgroup Under-Coverage in Alzheimer's Disease Longitudinal Prediction

Longitudinal prediction of Alzheimer's disease biomarkers increasingly informs clinical decisions, and a forecast is only useful if it also reports how much to trust it. Conformal prediction supplies this by wrapping any forecaster in a prediction band with a finite-sample coverage guarantee under exchangeability. However, standard population-level conformal prediction guarantees only marginal coverage and may mask substantial under-coverage within clinically important subgroups. We introduce a general mechanism-driven framework for auditing and repairing such subgroup under-coverage. Across two cohorts (ADNI, OASIS-3), two base forecasters, and nine attributes spanning genetic risk, demographics, and clinical severity, we find that population-level bands under-cover high-risk subgroups in 57 of 68 audited combinations, despite achieving nominal marginal coverage. We trace these failures to two mechanisms: (A) \emph{rarity}, where a group-conditional band calibrated on only $n$ patients covers at most $k/(n+1)$; and (B) \emph{tail-heaviness}, where a population-wide band is too narrow for a heavy-tailed subgroup and additional data cannot close the gap. Under-coverage falls disproportionately on patients with high genetic risk and disease severity (6.1 pp mean deficit, 95\% CI [3.3, 8.9]), while demographic groups remain at the target level on average (0.0 pp, CI [$-1.9$, 1.7]). We pair each mechanism with a corresponding conformal correction: cross-conformal pooling for rarity, per-subgroup calibration for tail-heaviness, and a coverage-safe marginal floor when both arise. Together, these corrections restore target coverage for nearly every high-risk subgroup across both cohorts and forecasters.

stat.ME↗

arXiv2026

LaMP: Learning Vision-Language-Action Policy with 3D Scene Flow as Latent Motion Prior

We introduce \textbf{LaMP}, a dual-expert Vision-Language-Action framework that embeds dense 3D scene flow as a latent motion prior for robotic manipulation.Existing VLA models regress actions directly from 2D semantic visual features, forcing them to learn complex 3D physical interactions implicitly.This implicit learning strategy degrades under unfamiliar spatial dynamics.LaMP addresses this limitation by aligning a flow-matching \emph{Motion Expert} with a policy-predicting \emph{Action Expert} through gated cross-attention.Specifically, the Motion Expert generates a one-step partially denoised 3D scene flow, and its hidden states condition the Action Expert without full multi-step reconstruction.We evaluate LaMP on the LIBERO, LIBERO-Plus, and SimplerEnv-WidowX simulation benchmarks as well as real-world experiments.LaMP consistently outperforms evaluated VLA baselines across LIBERO, LIBERO-Plus, and SimplerEnv-WidowX benchmarks, achieving the highest reported average success rates under the same training budgets. On LIBERO-Plus OOD perturbations, LaMP shows improved robustness with an average 9.7\% gain over the strongest prior baseline.Our project page is available at https://summerwxk.github.io/lamp-project-page/.

cs.CV↗

arXiv2026

A Surface-based Multimodal Framework for Multitask Analysis in Alzheimer's Disease

Alzheimer's Disease (AD) is a progressive neurodegenerative disorder, and longitudinal analysis is critical for early detection and effective intervention. Developing models capable of multimodal and multitask analysis enables a more comprehensive understanding of AD progression. However, multimodal learning remains challenged by cross-modal misalignment, non-Euclidean surface representations of cortical data, and limited data availability in small-sample clinical settings. In this work, we propose an augmented spherical data-driven multimodal framework for multitask AD analysis. A spherical diffusion model is first trained to generate paired cortical thickness and Tau PET Standardized Uptake Value Ratio (SUVR) data, enabling structurally consistent multimodal augmentation on cortical surfaces while preserving anatomical correspondence. The augmented data are subsequently used to train a contrastive learning model that learns aligned and fused cross-modal representations. This design strengthens multimodal integration and encourages more balanced representation learning. The learned imaging features are further integrated with tabular cognitive assessments and demographic variables, and processed using an in-context learning model to perform both classification and regression tasks without task-specific fine-tuning. Experiments on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset ($n = 802$) demonstrate consistent performance improvements across five diagnostic and longitudinal tasks, outperforming six baseline models.

eess.IV↗

arXiv2026

PCR: A Prefetch-Enhanced Cache Reuse System for Low-Latency RAG Serving

Retrieval-Augmented Generation (RAG) systems enhance the performance of large language models (LLMs) by incorporating supplementary retrieved documents, enabling more accurate and context-aware responses. However, integrating these external documents often results in very long input sequences, which significantly increases computation costs during the prefill stage, where key-value (KV) representations for all input tokens are generated. This latency bottleneck becomes especially pronounced under high-throughput serving scenarios. KV-cache reuse offers a promising solution by storing previously computed KV states for shared input prefixes, thereby avoiding redundant computation across requests that contain overlapping context. Yet, the effectiveness of cache reuse is often limited by three practical challenges: low cache hit rates due to naive eviction policies, high CPU-GPU data transfer overhead, and slow SSD I/O when caches spill to storage. To address these issues, we propose PCR, a system designed to maximize KV-cache reuse efficiency through intelligent prefetching and pipelined data movement. Specifically, PCR introduces three key techniques: (1) a prefix-tree caching structure with a look-ahead LRU replacement policy that uses pending requests in the scheduler queue to improve cache hit ratios; (2) layer-wise overlapping that pipelines KV-cache loading and GPU computation across CUDA streams to hide communication latency; and (3) queue-based prefetching that proactively loads relevant KV caches from SSD into DRAM before they are needed. Extensive experiments show that PCR outperforms existing KV-cache reuse methods, achieving up to a 2.47x speedup in terms of average TTFT.

cs.DC↗

arXiv2026

Synergizing Understanding and Generation with Interleaved Analyzing-Drafting Thinking

Unified Vision-Language Models (UVLMs) aim to advance multimodal learning by supporting both understanding and generation within a single framework. However, existing approaches largely focus on architectural unification while overlooking the need for explicit interaction between the two capabilities during task solving. As a result, current models treat understanding and generation as parallel skills rather than synergistic processes. To achieve real synergy, we introduce the interleaved Analyzing-Drafting problem-solving loop (AD-Loop), a new think paradigm that dynamically alternates between analytic and drafting operations. By interleaving textual thoughts with visual thoughts, AD-Loop enables models to iteratively refine both comprehension and outputs, fostering genuine synergy. To train this mechanism, we design a two-stage strategy: supervised learning on interleaved thought data to initialize alternation, followed by reinforcement learning to promote adaptive and autonomous control. Extensive experiments demonstrate that AD-Loop consistently improves performance across standard benchmarks for both understanding and generation, with strong transferability to various UVLMs architectures. Visual analyses further validate the effectiveness of implicit visual thoughts. These results highlight AD-Loop as a principled and broadly applicable strategy for synergizing comprehension and creation. The project page is at https://sqwu.top/AD-Loop.

cs.CV↗

arXiv2026

PolyJailbreak: Cross-Modal Jailbreaking Attacks on Black-Box Multimodal LLMs

Multimodal large language models (MLLMs) have become integral to a wide range of real-world applications by jointly reasoning over text and visual inputs. However, despite recent advances in safety alignment, MLLMs remain vulnerable to jailbreak attacks, where carefully crafted inputs can bypass safety mechanisms and elicit harmful responses. In this work, we investigate the security vulnerabilities of MLLMs in text-vision scenarios and propose a novel black-box jailbreak framework, named PolyJailbreak. We first identify a phenomenon, termed multimodal safety asymmetry, where visual alignment introduces uneven safety constraints across modalities and weakens overall robustness. We analyze attention dynamics and latent representations in MLLMs, revealing that visual inputs can disrupt cross-modal information flow and reduce the model's ability to separate benign and malicious intents. Motivated by these findings, we propose PolyJailbreak, which organizes the discovered vulnerabilities into a structured library of reusable Atomic Strategy Primitives to enable step-wise transformations from harmful intents to effective jailbreak inputs. Guided by these primitives, a reinforcement learning-based multi-agent optimization process automatically adapts attacks to the target model without access to internal parameters. Extensive experiments on a wide range of MLLMs demonstrate that PolyJailbreak consistently outperforms state-of-the-art jailbreak baselines, with an average improvement of 18.15% in attack success rate and a success rate exceeding 95% on commercial black-box models, including GPT-4o and Gemini.

cs.CR↗

arXiv2025

Impact of Target and Tool Visualization on Depth Perception and Usability in Optical See-Through AR

Optical see-through augmented reality (OST-AR) systems like Microsoft HoloLens 2 hold promise for arm's distance guidance (e.g., surgery), but depth perception of the hologram and occlusion of real instruments remain challenging. We present an evaluation of how visualizing the target object with different transparencies and visualizing a tracked tool (virtual proxy vs. real tool vs. no tool tracking) affects depth perception and system usability. Ten participants performed two experiments on HoloLens 2. In Experiment 1, we compared high-transparency vs. low-transparency target rendering in a depth matching task at arm's length. In Experiment 2, participants performed a simulated surgical pinpoint task on a frontal bone target under six visualization conditions ($2 \times 3$: two target transparencies and three tool visualization modes: virtual tool hologram, real tool, or no tool tracking). We collected data on depth matching error, target localization error, system usability, task workload, and qualitative feedback. Results show that a more opaque target yields significantly lower depth estimation error than a highly transparent target at arm's distance. Moreover, showing the real tool (occluding the virtual target) led to the highest accuracy and usability with the lowest workload, while not tracking the tool yielded the worst performance and user ratings. However, making the target highly transparent, while allowing the real tool to remain visible, slightly impaired depth cues and did not improve usability. Our findings underscore that correct occlusion cues, rendering virtual content opaque and occluding it with real tools in real time, are critical for depth perception and precision in OST-AR. Designers of arm-distance AR systems should prioritize robust tool tracking and occlusion handling; if unavailable, cautiously use transparency to balance depth perception and tool visibility.

cs.HC↗

arXiv2025

Interaction as Intelligence: Deep Research With Human-AI Partnership

This paper introduces "Interaction as Intelligence" research series, presenting a reconceptualization of human-AI relationships in deep research tasks. Traditional approaches treat interaction merely as an interface for accessing AI capabilities-a conduit between human intent and machine output. We propose that interaction itself constitutes a fundamental dimension of intelligence. As AI systems engage in extended thinking processes for research tasks, meaningful interaction transitions from an optional enhancement to an essential component of effective intelligence. Current deep research systems adopt an "input-wait-output" paradigm where users initiate queries and receive results after black-box processing. This approach leads to error cascade effects, inflexible research boundaries that prevent question refinement during investigation, and missed opportunities for expertise integration. To address these limitations, we introduce Deep Cognition, a system that transforms the human role from giving instructions to cognitive oversight-a mode of engagement where humans guide AI thinking processes through strategic intervention at critical junctures. Deep cognition implements three key innovations: (1)Transparent, controllable, and interruptible interaction that reveals AI reasoning and enables intervention at any point; (2)Fine-grained bidirectional dialogue; and (3)Shared cognitive context where the system observes and adapts to user behaviors without explicit instruction. User evaluation demonstrates that this cognitive oversight paradigm outperforms the strongest baseline across six key metrics: Transparency(+20.0%), Fine-Grained Interaction(+29.2%), Real-Time Intervention(+18.5%), Ease of Collaboration(+27.7%), Results-Worth-Effort(+8.8%), and Interruptibility(+20.7%). Evaluations on challenging research problems show 31.8% to 50.0% points of improvements over deep research systems.

cs.CL↗

arXiv2025

MRUCT: Mixed Reality Assistance for Acupuncture Guided by Ultrasonic Computed Tomography

Chinese acupuncture practitioners primarily depend on muscle memory and tactile feedback to insert needles and accurately target acupuncture points, as the current workflow lacks imaging modalities and visual aids. Consequently, new practitioners often learn through trial and error, requiring years of experience to become proficient and earn the trust of patients. Medical students face similar challenges in mastering this skill. To address these challenges, we developed an innovative system, MRUCT, that integrates ultrasonic computed tomography (UCT) with mixed reality (MR) technology to visualize acupuncture points in real-time. This system offers offline image registration and real-time guidance during needle insertion, enabling them to accurately position needles based on anatomical structures such as bones, muscles, and auto-generated reference points, with the potential for clinical implementation. In this paper, we outline the non-rigid registration methods used to reconstruct anatomical structures from UCT data, as well as the key design considerations of the MR system. We evaluated two different 3D user interface (3DUI) designs and compared the performance of our system to traditional workflows for both new practitioners and medical students. The results highlight the potential of MR to enhance therapeutic medical practices and demonstrate the effectiveness of the system we developed.

cs.HC↗

arXiv2024

DipMe: Haptic Recognition of Granular Media for Tangible Interactive Applications

While tangible user interface has shown its power in naturally interacting with rigid or soft objects, users cannot conveniently use different types of granular materials as the interaction media. We introduce DipMe as a smart device to recognize the types of granular media in real time, which can be used to connect the granular materials in the physical world with various virtual content. Other than vision-based solutions, we propose a dip operation of our device and exploit the haptic signals to recognize different types of granular materials. With modern machine learning tools, we find the haptic signals from different granular media are distinguishable by DipMe. With the online granular object recognition, we build several tangible interactive applications, demonstrating the effects of DipMe in perceiving granular materials and its potential in developing a tangible user interface with granular objects as the new media.

cs.HC↗