arXiv subjects

Ziqi Wang

Publications and source records attributed to Ziqi Wang.

At least 55 records · Page 3Linked to original sources

arXiv2026

GoLongRL: Capability-Oriented Long Context Reinforcement Learning with Multitask Alignment

We present GoLongRL, a fully open-source, capability-oriented post-training recipe for long-context reinforcement learning with verifiable rewards (RLVR). Existing long-context RL methods often treat data construction as a matter of designing increasingly complex retrieval paths, leading to homogeneous task coverage and reward formulations that inadequately reflect practical long-context requirements. Our work offers two contributions. (1) Capability-oriented data construction with full open release. We openly release a dataset of 23K RLVR samples, the complete construction pipeline, and all training code. Guided by a taxonomy of long-context capabilities, the dataset spans 9 task types, each paired with its natural evaluation metric. It comprises curated open-source samples from established corpora and synthetic samples whose QA pairs are generated from real source documents such as books, academic papers, and multi-turn dialogues. Under the same vanilla GRPO setup, our dataset alone outperforms the closed-source QwenLong-L1.5 dataset. Moreover, our Qwen3-30B-A3B model trained on this data delivers long-context performance comparable to DeepSeek-R1-0528 and Qwen3-235B-A22B-Thinking-2507, suggesting that broader coverage and greater reward diversity substantially benefit long-context capability improvement. (2) TMN-Reweight for heterogeneous multitask optimization. To address optimization challenges from heterogeneous rewards, we propose TMN-Reweight, which combines task-level mean normalization for cross-task reward scale alignment with difficulty-adaptive weighting for more reliable advantage estimation. TMN-Reweight further improves average performance over vanilla GRPO, with general capabilities preserved or improved across reported evaluations.

cs.CL↗

arXiv2026

InterLight: Leveraging Intrinsic Illumination Priors for Low-Light Image Enhancement

Low-Light Image Enhancement (LLIE) has long been a challenging problem in low-level vision, as insufficient illumination often leads to low contrast, detail loss, and noise. Recent studies show that deep learning-based Retinex theory can effectively decouple illumination and reflectance. However, existing methods frequently suffer from over-enhancement or color distortion, and often assume uniform noise or ideal lighting. To address these limitations, we propose InterLight, a novel framework that systematically excavates and operationalizes intrinsic illumination priors for LLIE.Our core insight is that robust enhancement requires not just estimating illumination, but constructing an illumination-aware pipeline. We first inject sensor-level illumination-response priors via physics-guided augmentation, then represent the degradation through adaptive prompts conditioned on the scene's latent illumination state. This explicit representation directly guides a luminance-gated intrinsic memory mechanism to selectively compensate for information loss, prioritizing reconstruction in dark regions while preserving fidelity in bright ones. Finally, the entire process is regularized by a self-supervised consistency objective that distills illumination-invariant features. By deeply exploiting intrinsic illumination priors, our method achieves clearer textures and more visually coherent enhancement results. Extensive experiments across multiple benchmarks demonstrate the effectiveness of our approach. Code is available at: https://github.com/House-yuyu/InterLight.

cs.CV↗

arXiv2026

Mixture-of-Experts Can Surpass Dense LLMs Under Strictly Equal Resource

Mixture-of-Experts (MoE) language models dramatically expand model capacity and achieve remarkable performance without increasing per-token compute. However, can MoEs surpass dense architectures under strictly equal resource constraints -- that is, when the total parameter count, training compute, and data budget are identical? This question remains under-explored despite its significant practical value and potential. In this paper, we propose a novel perspective and methodological framework to study this question thoroughly. First, we comprehensively investigate the architecture of MoEs and achieve an optimal model design that maximizes the performance. Based on this, we subsequently find that an MoE model with activation rate in an optimal region is able to outperform its dense counterpart under the same total parameter, training compute and data resource. More importantly, this optimal region remains consistent across different model sizes. Although additional amount of data turns out to be a trade-off for enhanced performance, we show that this can be resolved via reusing data. We validate our findings through extensive experiments, training nearly 200 language models at 2B scale and over 50 at 7B scale, cumulatively processing 50 trillion tokens. All model checkpoints are publicly available.

cs.CL↗

arXiv2026

Approximate and Weighted Data Reconstruction Attack in Federated Learning

Federated Learning (FL) is a distributed learning paradigm that enables multiple clients to collaborate on building a machine learning model without sharing their private data. Although FL is considered privacy-preserved by design, recent data reconstruction attacks demonstrate that an attacker can recover clients' training data based on the parameters shared in FL. However, most existing methods fail to attack the most widely used horizontal Federated Averaging (FedAvg) scenario, where clients share model parameters after multiple local training steps. To tackle this issue, we propose an interpolation-based approximation method, which makes attacking FedAvg scenarios feasible by generating the intermediate model updates of the clients' local training processes. Then, we design a layer-wise weighted loss function to improve the data quality of reconstruction. We assign different weights to model updates in different layers concerning the neural network structure, with the weights tuned by Bayesian optimization. Finally, experimental results validate the superiority of our proposed approximate and weighted attack (AWA) method over the other state-of-the-art methods, as demonstrated by the substantial improvement in different evaluation metrics for image data reconstructions.

cs.LG↗

arXiv2026

LoRA in LoRA: Towards Parameter-Efficient Architecture Expansion for Continual Visual Instruction Tuning

Continual Visual Instruction Tuning (CVIT) enables Multimodal Large Language Models (MLLMs) to incrementally learn new tasks over time. However, this process is challenged by catastrophic forgetting, where performance on previously learned tasks deteriorates as the model adapts to new ones. A common approach to mitigate forgetting is architecture expansion, which introduces task-specific modules to prevent interference. Yet, existing methods often expand entire layers for each task, leading to significant parameter overhead and poor scalability. To overcome these issues, we introduce LoRA in LoRA (LiLoRA), a highly efficient architecture expansion method tailored for CVIT in MLLMs. LiLoRA shares the LoRA matrix A across tasks to reduce redundancy, applies an additional low-rank decomposition to matrix B to minimize task-specific parameters, and incorporates a cosine-regularized stability loss to preserve consistency in shared representations over time. Extensive experiments on a diverse CVIT benchmark show that LiLoRA consistently achieves superior performance in sequential task learning while significantly improving parameter efficiency compared to existing approaches. The code is available at https://github.com/chanceche/LiLoRA.

cs.CV↗

arXiv2026

When Does Hierarchy Help? Benchmarking Agent Coordination in Event-Driven Industrial Scheduling

Recent advances in agent and multi-agent systems have shown strong performance on tool use, reasoning, and collaborative tasks. However, existing benchmarks mostly evaluate task completion in weakly coupled environments, and provide limited support for studying coordination in shared, dynamically evolving systems with hierarchy and coupled constraints. This leaves an important question underexplored: when do different coordination paradigms succeed or fail? We introduce Distributed Event-driven Scheduling Benchmark (DESBench), a benchmark for evaluating agent coordination in hierarchical event-driven scheduling. Built on a shared discrete-event driven environment in industrial scheduling, our benchmark captures multi-timescale decision making, partial observability, and dynamically coupled constraints. We define tasks and metrics that evaluate effectiveness, constraint alignment, coordination efficiency, and robustness, and focus on four representative coordination paradigms: centralized, hierarchical, heterarchical, and holonic. These paradigms correspond to distinct mechanisms of information flow, decision authority, and conflict resolution. Our controlled evaluations reveal clear coordination trade-offs: centralized coordination is robust and communication-efficient but scales poorly with difficulty; hierarchical coordination improves efficiency through decomposition but suffers from cross-level misalignment; heterarchical coordination is flexible but communication-heavy; and holonic coordination satisfies constraints well but loses global robustness. These findings demonstrate that coordination design fundamentally shapes agent system behavior in complex environments, revealing structural trade-offs that cannot be captured by outcome metrics alone and underscoring the imperative for more adaptive, principled, and dynamic coordination mechanisms in future MAS research.

cs.MA↗

arXiv2026

Trust or Abstain? A Self-Aware RAG Approach

Retrieval-augmented generation (RAG) improves large language models (LLMs) by incorporating external evidence, but it also introduces knowledge conflicts when retrieved contextual knowledge (CK) and parametric knowledge (PK) disagree or are both unreliable. Existing approaches mainly coordinate which source to use, without explicitly asking whether each answer path is correct. We argue that faithful RAG requires LLM self-awareness, namely the ability to recognize the limits of its own knowledge and reasoning. To ground this problem, we construct a model-specific, ground-truth-aligned knowledge-conflict benchmark by evaluating LLM backbones on PK-only and CK-conditioned answer paths over approximately 69K query-context instances per backbone, drawn from five conflict-QA datasets. We then introduce SABER, a Self-Aware Belief Estimator for RAG that requires no LLM fine-tuning. SABER combines a self-prior with PK-side and CK-side conditional reasoning representations from multi-trace inference, then estimates reliability beliefs with two lightweight predictors to drive a 4-cell decision over trust PK, trust CK, trust either, or abstain. Across four LLM backbones, SABER improves end-to-end accuracy and conflict-specific faithfulness over ten inference-time and fine-tuning baselines, with the largest gains on conflict-heavy datasets. Under abstention, SABER's risk-coverage curve Pareto-dominates every prompt-based abstainer, providing a tunable balance between coverage and answer risk. Our code is available at https://github.com/xizhu1022/SABER.

cs.IR↗

arXiv2026

Bifurcations in Stokes Flow Sedimentation

Particles whose shapes couple translation to rotation display a rich array of behaviors as they sediment at low Reynolds number. We introduce a unifying perspective in which the possible dynamical regimes and bifurcations between them can be understood. We use experimental measurements of helical ribbons, with controlled center of mass offsets, to identify the key bifurcation from complex dynamics to a single attracting state as the magnitude of the offset increases. The sedimentation dynamics are very sensitive to small center of mass offsets, with the bifurcation occurring for offsets less than one percent of the particle length. Using mobility tensors obtained from immersed boundary method simulations, we simulate helical particle sedimentation and identify an alignment bifurcation surface, defined in the three dimensional space of center of mass offsets, that separates simple from complex sedimentation dynamics. Inside this surface we find limit cycles which emerge through Hopf and homoclinic bifurcations. Cocentered particles with coincident centers of force and mobility provide a reference case at the center of the bifurcation surface. We show how the geometric and dynamical symmetries of sedimenting cocentered particles are broken as the center of force offset moves away from the cocentered case. Three parity time-reversal (PT) symmetries exist for all cocentered particles under reflections normal to the eigenvectors of its translation-rotation coupling tensor. When a center of force offset preserves at least one of these PT symmetries, then there are closed orbits for particles inside the alignment bifurcation surface.

physics.flu-dyn↗

arXiv2026

Reliable Microservice Tail Latency Prediction via Decoupled Dual-Stream Learning and Gradient Modulation

Microservice architectures enable scalable cloud-native applications; however, the distributed nature of these systems complicates the maintenance of strict Service Level Objectives. Accurately predicting window-level P95 tail latency remains difficult due to the complex interactions between software workload propagation and infrastructure resource limits. Existing predictive models struggle to capture these dynamics because the lack of explicit separation between traffic metrics and resource metrics causes misaligned feature representations. Building on this suboptimal data treatment, the unified architectures of prior approaches fail to isolate cascading service dependencies from localized processing capacity. Due to this entanglement, joint training suffers from an optimization imbalance wherein resource features converge faster and dominate gradient updates, thereby preventing the learning of underlying software topologies. To address these challenges, we propose USRFNet, a dual-stream framework that separates the modeling of demand and capacity. The proposed framework utilizes a Graph Neural Network to model the spatial interactions of traffic workloads across software-level service dependencies, and a gating MLP to independently extract infrastructure-level resource dynamics. The model then integrates these representations through hierarchical tensor fusion. To resolve the training imbalance, we introduce a Reliability-Aware Gradient Modulation strategy that dynamically rescales gradients based on the generalization ratio of each data stream. Experiments on three large-scale real-world benchmarks demonstrate that USRFNet outperforms state-of-the-art methods in prediction accuracy. Specifically, compared to the best-performing baselines, the proposed framework achieves relative MAPE reductions ranging from 15.62% to 26.11% across the evaluated datasets.

cs.LG↗

arXiv2026

Room-temperature third-order nonlinear anomalous Hall effect in ferromagnetic metal Fe3GaTe2

Berry curvature, as the imaginary component of quantum geometry, plays a crucial role in condensed matter physics. The spatial distribution of Berry curvature can be characterized by its dipole and multipole moments, which can induce the nonlinear anomalous Hall effect (NLAHE). To date, the NLAHE has been demonstrated in various materials, yet reports on room-temperature NLAHE are still limited. In this work, we report the observation of the third-order NLAHE in ferromagnetic metal Fe3GaTe2. The third-order NLAHE shows hysteretic behavior with the variation of magnetic field, where the coercive field is the same as that of the anomalous Hall effect, and the third-order NLAHE remains observable up to the Curie temperature (~350 K). The scaling analysis suggests that the third-order NLAHE may be attributed to the Berry curvature quadrupole. Our work not only provides an approach to study magnetic materials through nonlinear electric transports, but also opens up possibilities for the future development of room-temperature third-order nonlinear electronic devices.

cond-mat.mtrl-sci↗

arXiv2026

LoViF 2026 Challenge on Real-World All-in-One Image Restoration: Methods and Results

This paper presents a review for the LoViF Challenge on Real-World All-in-One Image Restoration. The challenge aimed to advance research on real-world all-in-one image restoration under diverse real-world degradation conditions, including blur, low-light, haze, rain, and snow. It provided a unified benchmark to evaluate the robustness and generalization ability of restoration models across multiple degradation categories within a common framework. The competition attracted 124 registered participants and received 9 valid final submissions with corresponding fact sheets, significantly contributing to the progress of real-world all-in-one image restoration. This report provides a detailed analysis of the submitted methods and corresponding results, emphasizing recent progress in unified real-world image restoration. The analysis highlights effective approaches and establishes a benchmark for future research in real-world low-level vision.

cs.CV↗

arXiv2026

Calibration-Aware Policy Optimization for Reasoning LLMs

Group Relative Policy Optimization (GRPO) enhances LLM reasoning but often induces overconfidence, where incorrect responses yield lower perplexity than correct ones, degrading relative calibration as described by the Area Under the Curve (AUC). Existing approaches either yield limited improvements in calibration or sacrifice gains in reasoning accuracy. We first prove that this degradation in GRPO-style algorithms stems from their uncertainty-agnostic advantage estimation, which inevitably misaligns optimization gradients with calibration. This leads to improved accuracy at the expense of degraded calibration. We then propose Calibration-Aware Policy Optimization (CAPO). It adopts a logistic AUC surrogate loss that is theoretically consistent and admits regret bound, enabling uncertainty-aware advantage estimation. By further incorporating a noise masking mechanism, CAPO achieves stable learning dynamics that jointly optimize calibration and accuracy. Experiments on multiple mathematical reasoning benchmarks show that CAPO-1.5B significantly improves calibration by up to 15% while achieving accuracy comparable to or better than GRPO, and further boosts accuracy on downstream inference-time scaling tasks by up to 5%. Moreover, when allowed to abstain under low-confidence conditions, CAPO achieves a Pareto-optimal precision-coverage trade-off, highlighting its practical value for hallucination mitigation.

cs.LG↗

arXiv2026

QuantVLA: Scale-Calibrated Post-Training Quantization for Vision-Language-Action Models

Vision-language-action (VLA) models unify perception, language, and control for embodied agents but face significant challenges in practical deployment due to rapidly increasing compute and memory demands, especially as models scale to longer horizons and larger backbones. To address these bottlenecks, we introduce QuantVLA, a training-free post-training quantization (PTQ) framework that, to our knowledge, is the first PTQ approach for VLA systems and the first to successfully quantize a diffusion transformer (DiT) action head. QuantVLA incorporates three scale-calibrated components: (1) a selective quantization layout that integerizes all linear layers in both the language backbone and the DiT while keeping attention projections in floating point to preserve the original operator schedule; (2) attention temperature matching, a lightweight per-head scaling mechanism that stabilizes attention logits and is folded into the dequantization scales at inference; and (3) output head balancing, a per-layer residual interface calibration that mitigates post-projection energy drift. The framework requires no additional training, uses only a small unlabeled calibration buffer, and supports integer kernels for low-bit weights and activations while leaving the architecture unchanged. Across representative VLA models on LIBERO, QuantVLA exceeds the task success rates of full-precision baselines, achieves about 70% relative memory savings on the quantized components, providing a practical pathway toward scalable low-bit embodied intelligence under strict compute, memory, and power constraints.

cs.LG↗

arXiv2026

Biologically Inspired Event-Based Perception and Sample-Efficient Learning for High-Speed Table Tennis Robots

Perception and decision-making in high-speed dynamic scenarios remain challenging for current robots. In contrast, humans and animals can rapidly perceive and make decisions in such environments. Taking table tennis as a typical example, conventional frame-based vision sensors suffer from motion blur, high latency and data redundancy, which can hardly meet real-time, accurate perception requirements. Inspired by the human visual system, event-based perception methods address these limitations through asynchronous sensing, high temporal resolution, and inherently sparse data representations. However, current event-based methods are still restricted to simplified, unrealistic ball-only scenarios. Meanwhile, existing decision-making approaches typically require thousands of interactions with the environment to converge, resulting in significant computational costs. In this work, we present a biologically inspired approach for high-speed table tennis robots, combining event-based perception with sample-efficient learning. On the perception side, we propose an event-based ball detection method that leverages motion cues and geometric consistency, operating directly on asynchronous event streams without frame reconstruction, to achieve robust and efficient detection in real-world rallies. On the decision-making side, we introduce a human-inspired, sample-efficient training strategy that first trains policies in low-speed scenarios, progressively acquiring skills from basic to advanced, and then adapts them to high-speed scenarios, guided by a case-dependent temporally adaptive reward and a reward-threshold mechanism. With the same training episodes, our method improves return-to-target accuracy by 35.8%. These results demonstrate the effectiveness of biologically inspired perception and decision-making for high-speed robotic systems.

cs.RO↗

arXiv2026

RAGRouter-Bench: A Dataset and Benchmark for Adaptive RAG Routing

Retrieval-augmented generation (RAG) has evolved into a family of paradigms with distinct performance profiles and resource demands, turning paradigm selection into a multi-criteria, context-dependent decision problem. Nevertheless, existing studies largely focus on isolated method improvements or query-only benchmarking, without systematically examining how RAG paradigms behave across diverse query-corpus contexts and effectiveness-efficiency trade-offs. In this work, we introduce RAGRouter-Bench, the first dataset and benchmark for adaptive RAG routing. Grounded in query-corpus compatibility, the benchmark integrates three canonical query types, fine-grained corpus indicators capturing structural and semantic properties, and a unified protocol for evaluating both generation quality and resource consumption. Then, we implement standardized RAG paradigms with multiple backbone LLMs across all query-corpus combinations, constructing a comprehensive benchmark with quantitative metrics and LLM-as-a-Judge evaluations to inform context-aware and cost-effective RAG routing decisions. We further formulate routing as context-dependent paradigm selection and benchmark a range of query-corpus routers on the constructed dataset. Extensive experiments demonstrate that no one-size-fits-all paradigm exists across query-corpus pairs, and that adaptive routing yields more favorable effectiveness-efficiency trade-offs than fixed paradigm selection. These findings establish query-corpus compatibility as a central principle for adaptive RAG routing and position RAGRouter-Bench as a systematic testbed for next-generation RAG systems.

cs.IR↗

arXiv2026

Social Amplification Dominates Collective Hazard Response

Large-scale hazards affect societies not only through direct physical impacts but also through emotions that spread across populations. Fueled by social amplification and networked communication, collective emotions often diverge markedly from underlying physical threats, pressuring policymakers toward suboptimal decisions that erode long-term societal resilience and misalign risk governance priorities. Yet when exactly these collective emotions mirror hazard severity and when they are warped by social dynamics remains poorly understood. We introduce a compact, interpretable model that couples hazard exposure with networked emotional contagion and identifies the transition from proportionate responses to an amplification regime sustained by negativity bias. Applying this framework to the COVID-19 pandemic in the United States, we integrate state-level epidemiological data with large-scale stress signals inferred from Twitter/X activity. Our analysis shows that social influence outweighed direct hazard forcing in over 80\% of U.S. states during the study period, and that amplified stress covaries with major economic indices. These findings reveal a measurable regularity in societal hazard response, enabling quantitative anticipation of collective emotional tipping points and supporting community resilience under large-scale hazards.

physics.soc-ph↗

arXiv2026

Story2Proposal: A Scaffold for Structured Scientific Paper Writing

Generating scientific manuscripts requires maintaining alignment between narrative reasoning, experimental evidence, and visual artifacts across the document lifecycle. Existing language-model generation pipelines rely on unconstrained text synthesis with validation applied only after generation, often producing structural drift, missing figures or tables, and cross-section inconsistencies. We introduce Story2Proposal, a contract-governed multi-agent framework that converts a research story into a structured manuscript through coordinated agents operating under a persistent shared visual contract. The system organizes architect, writer, refiner, and renderer agents around a contract state that tracks section structure and registered visual elements, while evaluation agents supply feedback in a generate evaluate adapt loop that updates the contract during generation. Experiments on tasks derived from the Jericho research corpus show that Story2Proposal achieved an expert evaluation score of 6.145 versus 3.963 for DirectChat (+2.182) across GPT, Claude, Gemini, and Qwen backbones. Compared with the structured generation baseline Fars, Story2Proposal obtained an average score of 5.705 versus 5.197, indicating improved structural consistency and visual alignment.

cs.CL↗

arXiv2026

RM-R1: Reward Modeling as Reasoning

Reward modeling is essential for aligning large language models with human preferences through reinforcement learning. To provide accurate reward signals, a reward model (RM) should stimulate deep thinking and conduct interpretable reasoning before assigning a score or a judgment. Inspired by recent advances of long chain-of-thought on reasoning-intensive tasks, we hypothesize and validate that integrating reasoning into reward modeling significantly enhances RM's interpretability and performance. We introduce a new class of generative reward models, Reasoning Reward Models (ReasRMs), which formulate reward modeling as a reasoning task. We propose a reasoning-oriented training pipeline and train a family of ReasRMs, RM-R1. RM-R1 features a chain-of-rubrics (CoR) mechanism -- self-generating sample-level chat rubrics or math/code solutions, and evaluating candidate responses against them. The training of RM-R1 consists of two key stages: (1) distillation of high-quality reasoning chains and (2) reinforcement learning with verifiable rewards. Empirically, our models achieve superior performance across three reward model benchmarks on average, outperforming much larger open-weight models (e.g., INF-ORM-Llama3.1-70B) and proprietary ones (e.g., GPT-4o) by up to 4.9%. Beyond final performance, we perform thorough analyses to understand the key ingredients of successful ReasRM training.

cs.CL↗