arXiv subjects

Wentao Qiu

Publications and source records attributed to Wentao Qiu.

10 recordsLinked to original sources

arXiv2026

SpecPrefetch: Parameter-Efficient Expert Prefetching for Sparse MoE Foundation Models

Sparse Mixture-of-Experts (MoE) models expand foundation model capacity through conditional expert activation, but their full expert pools remain difficult to deploy under limited accelerator memory. Although expert offloading alleviates memory pressure by moving inactive experts to host memory or storage, it introduces a routing-dependent transfer bottleneck: required experts are known only after native top-\(K\) routing, which serializes routing, expert loading, and expert execution during inference. To address this bottleneck, we propose SpecPrefetch, a parameter-efficient prefetching framework for offloaded MoE inference. SpecPrefetch uses a shared lightweight adapter to predict next-layer expert candidates only for asynchronous transfer, while the frozen native router still determines the final executed experts. By separating transfer prediction from execution routing, SpecPrefetch reduces exposed expert-loading latency without changing pretrained routing semantics, so prediction errors affect transfer efficiency rather than model outputs. In addition, a window-aware scheduler prioritizes feasible transfers under cache and bandwidth constraints. Across Qwen3-VL-30B-A3B and DeepSeek-VL2-Tiny, SpecPrefetch achieves the best average expert recall in 9 out of 10 model-benchmark settings with substantially fewer trainable parameters than learned predictor baselines. On a Snapdragon 8 Elite device, SpecPrefetch further improves decoding throughput by up to \(20\%\) over a compute-optimized offloading runtime, demonstrating practical benefits for storage-constrained MoE deployment. The code and model weights are available at https://github.com/wei390/SpecPrefetch.

cs.AI↗

arXiv2026

StepX-Edge: An On-Device UI Vision-Language Model via Architecture-Training-Deployment Co-Design

Deploying a vision-language model with full UI understanding on end devices has long been trapped between accuracy and efficiency: on one side is the accuracy bar for OCR, screen understanding, visual question answering, and element grounding; on the other is the strict compute, memory, and power budget of mobile chips. Existing work either trades one for the other, or stops at simulation without real-device validation. We present StepX-Edge, a 0.9B-parameter on-device UI vision-language model that resolves this tension through three-layer co-design of architecture, training, and deployment. Architecturally, UI-aware Layered Visual Encoding (ULVE) and a Progressive Dimensionality Projection (PDP) connector target the extreme aspect ratios and fine-grained perception of screens, while standard full attention throughout ensures native compatibility with mainstream mobile NPU operators. For training, the five-stage StepX-Curriculum framework is designed around our observation of mutual-promotion effects among UI subtasks, so that all four capabilities grow synergistically under a tight parameter budget rather than interfering. For deployment, a module-wise differentiated two-stage PTQ-to-QAT quantization scheme keeps the post-quantization accuracy loss within 1%. StepX-Edge achieves the strongest overall UI understanding among <=1B models, surpassing all 2B-2.3B baselines on ScreenQA (88.76 F1) and Chinese OCRBench v2 (57.25), and matching 1.3B-2.3B general VLMs on RefCOCO (92.0%) and OCRBench v1 (831) with far fewer parameters. After W4A16+KV8 quantization, the model runs stably on Snapdragon 8 Gen5 devices with ~0.84 s TTFT, 98 tok/s decode, and 1.4 GB peak memory. We will open-source the training data, the full training recipe, and the quantization deployment pipeline.

cs.CV↗

arXiv2026

ANCHOR: Abductive Network Construction with Hierarchical Orchestration for Reliable Probability Inference in Large Language Models

A central challenge in large-scale decision-making under incomplete information is estimating reliable probabilities. Recent approaches use Large Language Models (LLMs) to generate explanatory factors and coarse-grained probability estimates, which are then refined by a Naïve Bayes model over factor combinations. However, sparse factor spaces often yield ``unknown'' predictions, while expanding factors increases noise and spurious correlations, weakening conditional independence and degrading reliability. To address these limitations, we propose \textsc{Anchor}, an aggregated Bayesian inference framework over a hierarchical factor space. It constructs dense factor hierarchies through iterative generation and clustering, maps contexts via hierarchical retrieval and refinement, and augments Naïve Bayes with a Causal Bayesian Network to model latent factor dependencies. Experiments show that \textsc{Anchor} markedly reduces ``unknown'' predictions and produces more reliable probability estimates than direct LLM baselines, achieving state-of-the-art performance while significantly reducing time and token overhead.

cs.CL↗

arXiv2026

DimMem: Dimensional Structuring for Efficient Long-Term Agent Memory

Large language model (LLM) agents require long-term memory to leverage information from past interactions. However, existing memory systems often face a fidelity--efficiency trade-off: raw dialogue histories are expensive, while flat facts or summaries may discard the structure needed for precise recall. We propose \textbf{DimMem}, a lightweight dimensional memory framework that represents each memory as an atomic, typed, and self-contained unit with explicit fields such as time, location, reason, purpose, and keywords. This representation exposes the structure needed for dimension-aware retrieval, memory update, and selective assistant-context recall without storing full histories in the model context. Across LoCoMo-10 and LongMemEval-S, DimMem achieves \textbf{81.43\%} and \textbf{78.20\%} overall accuracy, respectively, outperforming existing lightweight memory systems while reducing LoCoMo per-query token cost by \textbf{24\%}. We further show that dimensional memory extraction is learnable by compact models: after fine-tuning on the DimMem schema, a Qwen3-4B extractor surpasses LightMem with GPT-4.1-mini on both benchmarks and reaches performance comparable to, or better than, much larger extractors in key settings. These results suggest that explicit dimensional structuring is an effective and efficient foundation for long-term memory in LLM agents. Code is available at https://github.com/ChowRunFa/DimMem.

cs.CL↗

arXiv2026

AGSC: Adaptive Granularity and Semantic Clustering for Uncertainty Quantification in Long-text Generation

Large Language Models (LLMs) have demonstrated impressive capabilities in long-form generation, yet their application is hindered by the hallucination problem. While Uncertainty Quantification (UQ) is essential for assessing reliability, the complex structure makes reliable aggregation across heterogeneous themes difficult, in addition, existing methods often overlook the nuance of neutral information and suffer from the high computational cost of fine-grained decomposition. To address these challenges, we propose AGSC (Adaptive Granularity and GMM-based Semantic Clustering), a UQ framework tailored for long-form generation. AGSC first uses NLI neutral probabilities as triggers to distinguish irrelevance from uncertainty, reducing unnecessary computation. It then applies Gaussian Mixture Model (GMM) soft clustering to model latent semantic themes and assign topic-aware weights for downstream aggregation. Experiments on BIO and LongFact show that AGSC achieves state-of-the-art correlation with factuality while reducing inference time by about 60% compared to full atomic decomposition.

cs.CL↗

arXiv2026

GCoT-Decoding: Unlocking Deep Reasoning Paths for Universal Question Answering

Chain-of-Thought reasoning can enhance large language models, but it requires manually designed prompts to guide the model. Recently proposed CoT-decoding enables the model to generate CoT-style reasoning paths without prompts, but it is only applicable to problems with fixed answer sets. To address this limitation, we propose a general decoding strategy GCoT-decoding that extends applicability to a broader range of question-answering tasks. GCoT-decoding employs a two-stage branching method combining Fibonacci sampling and heuristic error backtracking to generate candidate decoding paths. It then splits each path into a reasoning span and an answer span to accurately compute path confidence, and finally aggregates semantically similar paths to identify a consensus answer, replacing traditional majority voting. We conduct extensive experiments on six datasets covering both fixed and free QA tasks. Our method not only maintains strong performance on fixed QA but also achieves significant improvements on free QA, demonstrating its generality.

cs.CL↗

arXiv2026

DTCRS: Dynamic Tree Construction for Recursive Summarization

Retrieval-Augmented Generation (RAG) mitigates the hallucination problem of Large Language Models (LLMs) by incorporating external knowledge. Recursive summarization constructs a hierarchical summary tree by clustering text chunks, integrating information from multiple parts of a document to provide evidence for abstractive questions involving multi-step reasoning. However, summary trees often contain a large number of redundant summary nodes, which not only increase construction time but may also negatively impact question answering. Moreover, recursive summarization is not suitable for all types of questions. We introduce DTCRS, a method that dynamically generates summary trees based on document structure and query semantics. DTCRS determines whether a summary tree is necessary by analyzing the question type. It then decomposes the question and uses the embeddings of sub-questions as initial cluster centers, reducing redundant summaries while improving the relevance between summaries and the question. Our approach significantly reduces summary tree construction time and achieves substantial improvements across three QA tasks. Additionally, we investigate the applicability of recursive summarization to different question types, providing valuable insights for future research.

cs.CL↗

arXiv2020

Highly Efficient Second Harmonic Generation of Thin Film Lithium Niobate Nanograting near Bound States in the Continuum

Bound states in the continuum (BICs), a concept from quantum mechanics, are ubiquitous physical phenomena where waves will be completely locked inside physical systems without energy leaky. Such a physical phenomenon in optics will provide a platform for optical mode confinement to strengthen local field enhancement in nonlinear optics. Here we utilize an optical system consisting of asymmetric nanogratings and waveguide of thin film lithium niobate (LiNbO3) material to enhance second harmonic response near BICs. By breaking the symmetry of grating periodicity, we realize strong local field confined inside waveguide up to 25 times normalized to incident field (with dissymmetric factor of 0.2), allowing strong light-matter interaction in nonlinear material. From the numerical simulation, we theoretically demonstrate that such an optical system can greatly enhance second harmonic intensity enhancement of about 104 compared with undersigned LiNbO3 film and conversion efficiency reaching 1.53e-5 for dissymmetric factor=0.2 under illumination of 1.33 GW/(suqare cm). Surprisingly, we can predict that a giant enhancement of second harmonic conversion efficiency will exceed 8.13e-5 for dissymmetric factor=0.1 when the optical system is extremely close to BICs. We believe that such an optical system to trap local field inside is also accessible to promote the application of thin film lithium niobate in the field of integrated nonlinear optics.

physics.optics↗

arXiv2020

Optical anapole mode in nanostructured lithium niobate for enhancing second harmonic generation

Second harmonic generation (SHG) with a material of large transparency is an attractive way of generating coherent light sources at exotic wavelength range such as VUV, UV and visible light. It is of critical importance to improve nonlinear conversion efficiency in order to find practical applications in quantum light source and high resolution nonlinear microscopy, etc. Here an enhanced SHG with conversion efficiency up to the order of 0.01% at SH wavelength of 282 nm under 11 GW/cm2 pump power via the excitation of anapole in lithium niobite (LiNbO3, or LN) nanodisk through the dominating d33 nonlinear coefficient is investigated. The anapole has advantages of strongly suppressing far-field scattering and well-confined internal field which helps to boost the nonlinear conversion. Anapoles in LN nanodisk is facilitated by high index contrast between LN and substrate with properties of near-zero-index via hyperbolic metamaterial structure design. By tailoring the multi-layers structure of hyperbolic metamaterials, the anapole excitation wavelength can be tuned at different wavelengths. It indicates that an enhanced SHG can be achieved at a wide range of pump light wavelengths via different design of the epsilon-near-zero (ENZ) hyperbolic metamaterials substrates. The proposed nanostructure in this work might hold significances for the enhanced light-matter interactions at the nanoscale such as integrated optics.

physics.optics↗

arXiv2018

Casimir Thrust Force on a Rotating Chiral Particle

In the work, the thermal and vacuum fluctuation is predicted capable of generating a Casimir thrust force on a rotating chiral particle, which will push or pull the particle along the rotation axis. The Casimir thrust force comes from two origins: i) the rotation-induced symmetry-breaking in the vacuum and thermal fluctuation and ii) the chiral cross-coupling between electric and magnetic fields and dipoles, which can convert the vacuum spin angular momentum (SAM) to the vacuum force. Using the fluctuation dissipation theorem (FDT), we derive the analytical expressions for the vacuum thrust force in dipolar approximation and the dependences of the force on rotation frequency, temperature and material optical properties are investigated. The work reveals a new mechanism to generate a vacuum force, which opens a new way to exploit zero-point energy of vacuum.

quant-ph↗