arXiv subjects

Xinyu Wang

Publications and source records attributed to Xinyu Wang.

At least 55 records · Page 3Linked to original sources

arXiv2026

FlashAccel: Leveraging High-Bandwidth Flash (HBF) for High-Throughput LLM Inference

Large language model (LLM) inference is increasingly limited by the capacity of High-Bandwidth Memory (HBM) in GPUs, as model weights and KV cache grow rapidly. High-Bandwidth Flash (HBF) provides higher capacity than HBM while offering comparable bandwidth, making it a promising substrate for capacity-constrained LLM inference. However, its inherently high access latency, low bandwidth utilization, and lack of support for heterogeneous resource management make it difficult to integrate HBF into GPUs for LLM inference. We present FlashAccel, a co-designed system that enables efficient LLM inference using HBF. FlashAccel integrates HBF into HBM-based GPUs, providing architectural support to mitigate access latency. It improves bandwidth utilization through specialized data layouts for both model weights and KV cache, and introduces an HBF-aware storage management layer together with a programming model to organize persistent data in HBF and coordinate heterogeneous memory resources at the system level. Experimental results demonstrate that integrating six HBF stacks into the GPU enables FlashAccel to deliver an average improvement of 2.49$\times$ and 1.93$\times$ in throughput per GPU and energy efficiency over the HBM-only GPU under a 100ms latency constraint, respectively.

cs.AR↗

arXiv2026

Enhancing LLMs in Predictive Political QA with Semi-Structured Data

Predictive political question answering (QA), such as predicting how a political actor will vote, goes beyond factual lookup. External political resources offer rich historical evidence, but rarely contain the answer itself. Existing LLM augmentation methods, including actor-profile-based simulation and knowledge graph evidence injection, improve political reasoning but largely treat external resources as knowledge-based evidence, leaving prediction-relevant signals under-modeled. We identify two complementary signals for predictive political QA: actor stances that capture issue-specific preferences, and high-order structure signals that capture indirect dependencies among political actors. We propose PSL, a dual-view framework that converts semi-structured political records into inference-oriented evidence for LLMs. PSL extracts stance signals from question-relevant actor records in a semantic view, and learns structure-aware actor representations from an actor interaction graph in a vector view. Across three real-world datasets and multiple LLMs, PSL consistently outperforms baselines, with ablations confirming the complementary gains of stance and structure signals.

cs.AI↗

arXiv2026

Alignment Drift in Single-Model Speculative Decoding for ASR: Mechanism, Correction, and Cost

Speculative decoding speeds up generation by letting a cheap draft propose several tokens that a target model checks in one pass. In the single-model form, the draft is a lightweight module attached to the target rather than a separate model. Applying this design to Automatic Speech Recognition (ASR) introduces an extra problem. The draft can read the whole audio at every step, yet its proposals get worse as it runs on its own. Access is not localization. The accepted text keeps the transcript position explicit, but the draft must also track the changing audio position. In the primary matched comparison, per-step audio access changes the first proposal modestly but roughly doubles later-proposal acceptance. Fixed-width windows show that the audio position explains part of this gap. A correctly placed window recovers continuation, while an equally narrow window at the wrong position reduces it. Late-draft median error reaches 21 frames in the hardest reported condition, while target attention during verification stays within a 2-frame median. We test two ways to reduce this drift. The first reads the audio position from verification attention and uses it to guide the next draft round. It saves time only when the extra accepted tokens offset the readout cost. The second is AnchorDraft, which teaches the draft to track the audio position during training without changing the inference graph. The trained draft improves end-to-end speed at both tested target scales. These results show that ASR self-speculation depends on token prediction, audio-position tracking, and draft cost.

cs.SD↗

arXiv2026

P2MFDS: A Privacy-Preserving Multimodal Fall Detection System for Elderly People in Bathroom Environments

By 2050, people aged 65 and over are projected to make up 16% of the global population. As aging is closely associated with increased fall risk, particularly in wet and confined environments such as bathrooms where over 80% of falls occur. Although recent research has increasingly focused on non-intrusive, privacy-preserving approaches that do not rely on wearable devices or video-based monitoring, these efforts have not fully overcome the limitations of existing unimodal systems (e.g., WiFi-, infrared-, or mmWave-based), which are prone to reduced accuracy in complex environments. These limitations stem from fundamental constraints in unimodal sensing, including system bias and environmental interference, such as multipath fading in WiFi-based systems and drastic temperature changes in infrared-based methods. To address these challenges, we propose a Privacy-Preserving Multimodal Fall Detection System for Elderly People in Bathroom Environments. First, we develop a sensor evaluation framework to select and fuse millimeter-wave radar with 3D vibration sensing, and use it to construct and preprocess a large-scale, privacy-preserving multimodal dataset in real bathroom settings, which will be released upon publication. Second, we introduce P2MFDS, a dual-stream network combining a CNN-BiLSTM-Attention branch for radar motion dynamics with a multi-scale CNN-SEBlock-Self-Attention branch for vibration impact detection. By uniting macro- and micro-scale features, P2MFDS delivers significant gains in accuracy and recall over state-of-the-art approaches.

cs.CV↗

arXiv2026

Quantization-Aware Neuromorphic Architecture for Skin Lesion Classification on Resource-Constrained Devices

On-device skin lesion analysis is constrained by the compute and energy cost of conventional CNN inference and by the need for lightweight calibration under clinical data shift. Neuromorphic processors provide event-driven sparse computation, but practical deployment is often limited by CNN-to-SNN conversion failures, including unsupported operators, quantization distortion, and accuracy degradation under class imbalance. We propose QANA, a quantization-aware CNN backbone within an end-to-end pipeline for conversion-stable neuromorphic execution. QANA improves conversion robustness by bounding intermediate activations, aligning normalization with low-bit quantization, and replacing conversion-fragile components with spike-compatible transformations. Efficient representation is achieved through Ghost-based feature generation, spatially-aware efficient channel attention, and squeeze-and-excitation modules whose operations can be quantized, folded, or lowered into integer graph operations. The quantized projection head produces SNN-ready logits and supports lightweight readout calibration without full retraining or data offloading. On HAM10000, QANA achieves 91.6% Top-1 accuracy and 91.0% macro F1, improving the strongest converted SNN baseline by 3.5 points in accuracy and 12.1 points in macro F1. On a clinical dataset, QANA achieves 90.8% Top-1 accuracy and 81.7% macro F1, improving the strongest baseline by 3.2 points in accuracy and 3.6 points in macro F1. On BrainChip Akida, QANA runs in 1.5 ms with 1.7 mJ per image, outperforming the strongest size-matched Akida baseline by 4.1 points in accuracy, 10.0 points in macro F1, 21.1% in latency, and 22.7% in energy.

eess.IV↗

arXiv2026

LiDAR-based 3D Change Detection at City Scale

High-definition 3D city maps enable city planning and change detection, which is essential for municipal compliance, map maintenance, and asset monitoring, including both built structures and urban greenery. Conventional Digital Surface Model (DSM) and image differencing are sensitive to vertical bias and viewpoint mismatch, while original point cloud or voxel models require large memory, assume perfect alignment, and degrade thin structures. We propose an uncertainty-aware, object-centric method for city-scale LiDAR-based change detection. Our method aligns data from different time periods using multi-resolution Normal Distributions Transform (NDT) and a point-to-plane Iterative Closest Point (ICP) method, normalizes elevation, and computes a per-point level of detection from registration covariance and surface roughness to calibrate change decisions. Geometry-based associations are refined by semantic and instance segmentation and optimized using class-constrained bipartite assignment with augmented dummies to handle split-merge cases. Tiled processing bounds memory and preserves narrow ground changes, while instance-level decisions integrate overlap, displacement, and volumetric differences under local detection gating. We perform experiments on the city of Subiaco, Western Australia, using datasets captured in 2023 and 2025. Our method achieves 95.3% accuracy, 90.8% macro F1, and 82.9% macro IoU, improving over the strongest baseline, Triplet KPConv, by 0.3, 0.6, and 1.1 percentage points, respectively.

cs.CV↗

arXiv2026

LoDA: A Level of Detection Aware Method and a Multimodal Sensing Benchmark for Object Level Change Detection

High-definition 3D LiDAR maps are important for autonomous driving and smart-city services, which require reliable detection of object-level changes in multi-temporal urban LiDAR to keep digital maps aligned with the physical world. Existing approaches from raster height differencing to depth image and point-cloud networks often remain tile-based and threshold-driven, yielding per-point scores without explicit detection limits or consistent object-level labels. We propose an object-level 3D change-detection pipeline that integrates detection-limit-aware registration, geometry-driven object proxies with rule-based semantic and instance segmentation, and displacement cues in height, volume, and surface-normal direction to assign five change labels with confidence. By decoupling registration, geometry, and semantics, the pipeline propagates pose uncertainty into spatially varying detection limits, stabilizes cross-epoch correspondences, and suppresses false changes caused by residual misalignment and density variation. We also present LoDA, a level-of-detection (LoD) aware benchmark for the Subiaco district with fused multi-temporal vehicle-LiDAR maps constructed with LiDAR, GNSS, and IMU support, semantic instances, and object-level annotations. On this benchmark, our method achieves 95.0% accuracy, 90.8% macro F1, and 83.0% macro IoU, exceeding the best baseline by 8.7 IoU points and 4.4 F1 points. On the public Urb3DCD-V2 benchmark evaluated under the official point-wise protocol, it reaches 96.81% mean accuracy and 89.52% mean change IoU, improving over the strongest reported baselines by 1.36 points in mAcc and 3.18 points in mIoUch.

cs.CV↗

arXiv2026

TreeSpark: Calibrated, Load-Adaptive Draft Trees for Semi-Autoregressive Speculative Decoding

Speculative decoding accelerates language-model inference by letting a cheap drafter propose tokens that the target model verifies in parallel. Recent block drafters make drafting nearly free: a single backbone pass emits an entire block of draft tokens. Draft trees promise a further gain -- several alternative continuations verified in one target forward -- but existing constructions rank candidates by per-position marginals that ignore which parent a candidate extends, so on semi-autoregressive drafters wider trees mostly add mis-ranked nodes; and a tree of fixed size ignores how much speculation each decoding round, and each serving load, can support. We introduce TreeSpark, which reads a parent-conditioned distribution from the drafter's existing Markov head at negligible cost, calibrates it into an edge-acceptance estimate, and lets path survival govern everything else: best-first expansion, per-round stopping, and a load-adaptive serving policy. Sampling siblings without replacement, with matching residuals in recursive rejection, keeps decoding lossless at any temperature. Adaptive trees improve on matched fixed budgets at every temperature; against a tuned chain on the same drafter, TreeSpark accepts 15-25% more draft tokens per round and decodes 8-14% faster in single-request wall-clock, and under rising load it gracefully shrinks the tree back to the chain. Code and artifacts: https://github.com/PopSoda2002/TreeSpark

cs.CL↗

arXiv2026

HoloQ-VLA: Uniform W4A4 Quantization of Vision-Language-Action Models

Vision-Language-Action (VLA) models unify perception, reasoning, and control in a single policy, but their multi-billion-parameter backbones and diffusion-based action heads make on-device deployment prohibitively expensive. Low-bit post-training quantization (PTQ) is the natural remedy, yet the diffusion action head that emits continuous control signals is highly sensitive to it: a few weight and activation outliers are enough to destabilize the head, so prior work leaves it at full precision or falls back to mixed-precision schemes, and uniformly quantizing the whole model to low bit-width remains an open challenge. We present HoloQ-VLA, the first training-free PTQ framework that compresses both the language backbone and the entire diffusion action head to uniform W4A4 precision without mixed-precision allocation. Instead of trading weight quality against activation quality, HoloQ-VLA targets the two outlier sources with complementary transforms: a weight-adapted rotation composed with an activation-dispersing Hadamard transform, together with per-step scaling that absorbs the dynamic-range drift exhibited by the action head across denoising steps. On LIBERO, HoloQ-VLA compresses Pi-0.5 and GR00T-N1.5 to W4A4 with 98.0% and 87.8% task success rates, matching or exceeding their FP16 references of 97.1% and 87.0%, while reducing the static memory footprint by 74.2%. Real-world manipulation experiments further demonstrate that HoloQ-VLA maintains smooth and accurate control across diverse real-world scenarios.

cs.CV↗

arXiv2026

DUET: A Diversity-Quality Duet of Distillation Experts for Two-Step Video Generation

Diffusion models have enabled high-quality video generation in recent years, but the high cost of iterative sampling hinders their practical deployment. Few-step distillation alleviates this cost, yet exposes a quality--diversity trade-off between its two dominant paradigms: trajectory-level distillation (e.g., sCM) favors diversity, whereas distribution-level distillation (e.g., DMD) favors quality. Targeting extreme two-step video generation, we introduce DUET, which reconciles the two paradigms through a noise-level duet of experts: an sCM expert takes the high-noise step to lay out diverse structure, and a DMD expert takes the low-noise step to refine appearance detail. Since the two experts are trained independently with their native objectives, DUET sidesteps the optimization difficulties of loss-level combinations and delivers quality and diversity jointly rather than trading one for the other. We further identify the relay interface and the high-noise stage as the remaining bottlenecks, and address them with RL-guided expert adaptation, yielding DUET+. With the Wan2.1-T2V-1.3B backbone, DUET lifts the two-step quality of sCM close to the level of DMD while retaining nearly all of its structural diversity---about twice that of DMD---and DUET+ further improves overall quality while preserving this diversity advantage. Together, these results establish noise-level expert specialization as a simple, effective paradigm for reconciling diversity and quality in two-step video generation.

cs.CV↗

arXiv2026

Diagnostic-Driven Layer-Wise Compensation for Post-Training Quantization of Encoder-Decoder ASR Models

Layer-wise post-training quantization reconstructs each layer from inputs already altered by the quantized prefix. QEP compensates for this drift with one model-wide coefficient, conflating the model-level operating point with residual variation across layers. We present FADE, which constructs layer-specific coefficients from normalized round-to-nearest distortion and a heuristic calibrated-solver response. It requires no training or per-model coefficient search and adds no inference-time operation. We evaluate seven Whisper, Moonshine, and Qwen3-ASR models at 3 and 4 bits on four English ASR benchmarks. Across 38 settings, FADE lowers mean word error rate relative to fixed QEP-0.5 in 31, although a development-tuned global coefficient recovers much of this gap. The largest absolute reductions occur in 3-bit settings whose final error remains too high for practical use; these results measure collapse mitigation rather than deployable accuracy. In two lower-WER 3-bit cases, FADE reduces the tuned-global result from 3.67 to 3.10 and from 13.03 to 11.63. Most 4-bit differences from the tuned control are within a descriptive tolerance, and one reverses. Paired reruns and within-seed permutations on an outcome-informed subset support assignment sensitivity in selected cases, but do not estimate a matrix-wide success rate. FADE is therefore a layer-wise alternative when per-model coefficient search is unavailable, not a universal replacement for tuned global compensation.

cs.SD↗

arXiv2026

InfMem: Learning System-2 Memory Control for Long-Context Agent

Reasoning over ultra-long documents requires synthesizing sparse evidence scattered across distant segments under strict memory constraints. While streaming agents enable scalable processing, their passive memory update strategy often fails to preserve low-salience bridging evidence required for multi-hop reasoning. We propose InfMem, a control-centric agent that instantiates System-2-style control via a PreThink-Retrieve-Write protocol. InfMem actively monitors evidence sufficiency, performs targeted in-document retrieval, and applies evidence-aware joint compression to update a bounded memory. To ensure reliable control, we introduce a practical SFT-to-RL training recipe that aligns retrieval, writing, and stopping decisions with end-task correctness. On ultra-long QA benchmarks ranging from 32k to 1M tokens, InfMem consistently outperforms MemAgent across backbones. Specifically, InfMem improves average absolute accuracy by 10.17, 11.84, and 8.23 points on Qwen3-1.7B, Qwen3-4B, and Qwen2.5-7B, respectively, while reducing inference time by 3.9x on average and by up to 5.1x through adaptive early stopping. Code is available at https://github.com/UCMP13753/InfMem.

cs.CL↗

arXiv2026

CARE: Pre-Execution Command Verification for Shell-Executing LLM Agents

Large Language Model (LLM) agents are increasingly used for coding and terminal automation, making shell-command dispatch a high-stakes runtime control point. We study command-level pre-execution mediation for individual shell commands produced by LLM agents under bounded path context. Existing safeguards remain limited: generic guardrails do not model shell structure in sufficient detail, always-on LLM judges are relatively costly and variable, and shell parsers do not directly prevent harmful execution. We present CARE (Canonicalization, Attribution, and Resolution Engine), a shell-specific, static-first verifier for individual shell commands before execution. CARE canonicalizes generated commands into stable verification targets, derives deterministic evidence over syntax, command semantics, path context, and provenance-backed risk patterns, and escalates only underdetermined cases to an LLM judge. This design keeps the common case fast, reproducible, and auditable while reserving neural adjudication for borderline commands. On the balanced main split, CARE reaches 85.64% F1 with a 0.91% false-positive rate at 2.32 ms mean latency. When deployed in its static enforcement profile, CARE retains 84.99% F1 at 0.34 ms and reduces realised harm on RedCode-gen to 37.33%. Across external-generalization tests and controlled Docker-sandbox execution, these profiles expose a practical trade-off between benign recovery, false-positive burden, latency, and harm reduction. Overall, command-level shell mediation can reduce dispatch-boundary risk for LLM agents while preserving most benign workflows.

cs.CR↗

arXiv2026

Statistical Verification of Medium-Access Parameterization for Power-Grid Edge Ad Hoc Sensor Networks

The widespread deployment of power grid ad hoc sensor networks based on IEEE 802.15.4 raises reliability challenges when nodes selfishly adapt CSMA/CA parameters to maximize individual performance. Such behavior degrades reliability, energy efficiency, and compliance with strict grid constraints. Existing analytical and simulation approaches often fail to rigorously evaluate configurations under asynchronous, event-driven, and resource-limited conditions. We develop a verification framework that integrates stochastic timed hybrid automata with statistical model checking (SMC) with confidence bounds to formally assess CSMA/CA parameterizations under grid workloads. By encoding node- and system-level objectives in temporal logic and automating protocol screening via large-scale statistical evaluation, the method certifies Nash equilibrium strategies that remain robust to unilateral deviations. In a substation-scale scenario, the certified equilibrium improves utility from 0.862 to 0.914 and raises the delivery ratio from 89.5% to 93.2% when compared with an aggressive tuning baseline. Against a delivery-oriented baseline, it reduces mean per-cycle energy from 152.8 mJ to 149.2 mJ while maintaining comparable delivery performance. Certified configurations satisfy latency, reliability, and energy constraints with robustness coefficients above 0.97 and utility above 0.91.

cs.NI↗

arXiv2026

Complete strategy spaces reveal hidden pathways to cooperation

Understanding how cooperation emerges and persists is a central challenge in evolutionary game theory. Existing models often rely on restricted, hand-picked strategy sets, which can overlook critical behavioural pathways. A recent four-strategy framework showed that cheap talk can promote cooperation through local interactions, yet it remained unclear whether modelled strategies might alter these conclusions. Here, we extend this framework to the complete set of eight strategies that naturally arise from communication and decision-making rules. We show that incorporating the full strategy space dramatically changes the evolutionary landscape. Cooperation becomes both more robust and more versatile, driven by novel pathways absent in the restricted model. In particular, we uncover a previously overlooked mechanism in which suspicious cooperation catalyses a cyclic dynamic that sustains cooperation. Conversely, the assumed role of strategic defection in the biased model is fragile, acting mainly as a spoiler rather than a genuine evolutionary attractor. The complete model further reveals a rich spectrum of long-term behaviours, including stable coexistence among up to seven strategies and time-varying patterns of partial coexistence. These results demonstrate that the full strategy space unlocks hidden routes to cooperative behaviour and highlight the importance of comprehensive modelling when explaining the emergence of cooperation.

q-bio.PE↗

arXiv2026

LiveLight: Real-time Streaming Video Relighting with Interactive Control

We present LiveLight, the first diffusion-based framework for real-time streaming video relighting with interactive 3D lighting control. Achieving this is non-trivial, as it requires overcoming three critical challenges: effectively injecting dynamic 3D lighting into a diffusion model, maintaining high-fidelity generation under an extremely low NFE (Number of Function Evaluations) budget for real-time speed, and facilitating continuous streaming for interactive control. To address these pain points, we propose three key designs. First, for accurate lighting injection, we propose a lightweight adapter that feeds Multi-Plane Light Irradiance (MPLI) conditions-depth-aware irradiance maps encoding 3D lighting geometry-directly into the diffusion backbone. Second, to prevent rendering quality degradation at low NFEs towards real-time distillation, we introduce a geometry-guided feedback branch. This training-time constraint leverages a frozen geometry estimator to enforce depth- and normal-consistent relighting, ensuring geometrically plausible shading without adding inference overhead. Finally, to enable streaming interaction, we develop a progressive rolling-window strategy that maintains a denoising ladder of latent chunks at varying noise levels. By propagating intermediate states, this strategy guarantees temporal coherence and supports arbitrarily long video relighting with per-frame reference refresh. Extensive experiments on real-world and synthetic benchmarks demonstrate that LiveLight achieves state-of-the-art relighting quality while running at real-time speed, significantly outperforming offline baselines in temporal stability, lighting controllability, and user preference. To foster real-time interactive relighting research, we will publicly release our models, training data, and synthetic data generator.

cs.CV↗

arXiv2026

Mapping melliferous tree species in Kenya via one-class classification with hyperspectral unsupervised domain adaptation

The beekeeping sector holds significant potential for livelihood diversification among the agropastoral communities in Kenya. Melliferous tree species play a critical role by providing essential nectar sources for bees. However, limited knowledge of their precise spatial distributions constrains the full development of beekeeping. One-class classification (OCC) offers a practical solution for detecting single target species without requiring extensive labeled data from other classes. Although existing OCC methods perform well in trained domains, the generalization capability to unseen domains remains limited due to domain shift. To address these challenges, this study proposes a hyperspectral unsupervised domain adaptation OCC framework (HyUDA-One) for tree species mapping using airborne hyperspectral imagery and laser scanning data. The spatial-spectral regularized pseudo-positive learning was designed to mitigate domain shift and improve model generalizability. The effectiveness of HyUDA-One was demonstrated by mapping three key melliferous tree species in two savanna landscapes in southern Kenya. The results show that HyUDA-One significantly improves performance in unlabeled domains. The F1-scores of 0.788, 0.845, and 0.768 were achieved for Senegalia mellifera, Vachellia tortilis, and Commiphora africana in the trained domain, respectively. In the untrained domain, the F1-scores of Senegalia mellifera and Vachellia tortilis were 0.756 and 0.884, respectively. The distribution maps revealed the spatial patterns of these melliferous tree species and the nectar source availability, offering an important reference for sustainable beekeeping development in savanna landscapes. Furthermore, the proposed framework can potentially be extended to other mapping applications, such as invasive species detection.

eess.IV↗

arXiv2026

SIGMA: Semantic Identifier Grouping for Molecular Autoregression

Autoregressive molecular models assign probability to molecular serializations even though chemical identity is invariant to serialization. Equivalent serializations can therefore represent a common molecular identity yet induce inconsistent next-token decisions. Randomized strings broaden exposure, but do not reveal which intermediate decisions should agree. We introduce SIGMA, a dense suffix-position objective built from chemically certified same-suffix triplets: two equivalent histories, one non-equivalent history, and a shared suffix. SIGMA aligns corresponding pre-token hidden states along the continuation and separates the negative to a finite relative margin, leaving the language-model objective, decoder, and inference procedure unchanged. We compare SIGMA with canonical training, randomized-serialization training, and last-token alignment across four datasets under SMILES and SELFIES. Across the eight representation-dataset blocks, SIGMA yields clear test-reference Frechet ChemNet Distance reductions in six: all four SELFIES domains and QM9 and ZINC under SMILES, with paired 95% confidence intervals below zero against every control. Position-wise analyses show improved state correspondence, chemical discrimination, and next-token agreement while preserving between-molecule information. On two full-corpus ZINC blocks, compute-matched ablations identify chemically correct state correspondence as the effective ingredient. Beyond generation, SIGMA improves mean predictive performance on all six molecular property benchmarks and reduces sensitivity to equivalent molecular serializations on every task.

cs.LG↗