Searcharxiv⌕ Search

arXiv subjects

Zijing Wang

Publications and source records attributed to Zijing Wang.

At least 19 recordsLinked to original sources
arXiv2026

A Target-Centric Survey of Quantization-Aware Training

The rapid development of LLMs incurs prohibitive memory footprints and intensive computational demands. Quantization-Aware Training (QAT) techniques have emerged as a promising solution to address these challenges by explicitly simulating quantization effects during model training, yielding low-bit models that achieve accuracy comparable to their full-precision counterparts. In this work, we provide a target-centric survey of QAT, aimed at clarifying both its theoretical foundations and its evolving implementation landscape. We systematically review existing QAT methods through a target-centric taxonomy and synthesize cross-target differences in error characteristics, numerical formats, and strategy transferability. We further summarize QAT evaluation paradigms and discuss challenges in optimization and deployment, outlining potential directions for future research.

cs.LG↗
arXiv2026

TRAM: Enhancing Multimodal Reasoning with Trajectory-Derived Auxiliary Memory

Multimodal Large Reasoning Models (MLRMs) have achieved strong performance on tasks requiring visual understanding and multi-step inference. However, as reasoning trajectories grow, models may become less effective at using information established earlier in the context, increasing the risk of reasoning errors. Existing approaches primarily address this problem by sustaining visual grounding throughout reasoning. However, reasoning also transforms visual observations into task-specific relations, constraints, and intermediate conclusions whose influence may weaken over long trajectories. Our attribution analysis suggests that correctness is not consistently separated by image attribution alone, but is more closely associated with whether trajectories retain and integrate such reasoning-derived information across stages. Motivated by this, we introduce TRAM (TRajectory-derived Auxiliary Memory), a training-free method that augments standard decoding with an auxiliary memory pathway derived from the model's own reasoning trajectory. TRAM consolidates completed reasoning into a compact latent memory, updates it online through fast and slow recurrent streams, and feeds it back into selected decoder layers through a lightweight residual pathway. Experiments across four MLRM variants on eight benchmarks show that TRAM improves performance over vanilla decoding on mathematical, scientific, and general visual reasoning tasks without additional training.

cs.CL↗
arXiv2026

An Information-Theoretic Metric for Semantic Value of Spatiotemporal Information

With the explosive growth of network scale and data volume, wireless communication is facing an increasingly severe limitation of spectrum resources. Semantic communication has emerged as a promising paradigm to break the bandwidth bottleneck by transmitting significant task-oriented semantic information rather than raw data. In practical real-time wireless applications, semantics of information exhibit diverse spatial and temporal correlations depending on intrinsic dynamics of source and extrinsic dynamics of environment. Motivated by this observation, this paper develops a novel information-theoretic metric to quantify the semantic value of spatiotemporal information. Specifically, a semantic value of information (SVoI) framework is proposed based on the mutual information, which characterises the reduction in uncertainty when predicting an unknown system state using past semantic spatiotemporal correlated observations. Focusing on general Gaussian Markov models, closed-form expressions of the SVoI are derived. Effects of both separable and coupled spatiotemporal correlations on SVoI are further investigated analytically. Numerical simulations are conducted to validate the theoretical analysis of SVoI and its bounds. The proposed SVoI metric jointly captures the impact of semantic spatiotemporal correlation of source, timeliness of information, and channel conditions, which could serve as an effective optimisation objective for the design of next-generation semantic-aware communication systems.

cs.IT↗
arXiv2026

Beyond Text Following: Repairable Arbitration Reversals in Audio-Language Models

Audio-language models (ALMs) often follow text that conflicts with audio, even when the audio evidence is clear. This raises a basic question: is the audio-supported answer unavailable, or is it represented but overridden by the conflicting text? We examine this question using a same-audio counterfactual that keeps the audio fixed, removes only the conflicting text, and measures the resulting shift in model preference. Across five ALMs and four conflict tasks, 64.1% of conflict samples show a sign flip: the same-audio branch prefers the audio-supported answer, whereas the joint branch prefers the text-supported answer. This pattern suggests that the relevant audio evidence is encoded but loses in arbitration. Activation patching further localizes the reversal to answer-position computation, and patching effects closely track output candidate-score differences (Spearman rho=0.93). Using this diagnostic, we propose Gated Audio Counterfactual Logit Correction (GACL), a training-free decoding rule that interpolates between joint and same-audio scores. Under a strict 5 pp faithfulness-drop budget, GACL improves nAUC by 17.8 points over the best contrastive baseline and transfers without retuning to vision-text arbitration (up to +40.5 pp).

cs.SD↗
arXiv2026

SMoA: Spectrum Modulation Adapter for Parameter-Efficient Fine-Tuning

As the number of model parameters increases, parameter-efficient fine-tuning (PEFT) has become the go-to choice for tailoring pre-trained large language models. Low-rank Adaptation (LoRA) uses a low-rank update method to simulate full parameter fine-tuning, which is widely used to reduce resource requirements. However, decreasing the rank encounters challenges with limited representational capacity. Theory suggests that LoRA fine-tuning with rank r converges toward the top r singular values of the pre-trained weight matrix. As the rank increases, more principal singular directions are preserved, which generally improves the model's performance. However, a larger rank also introduces more trainable parameters, leading to higher computational cost. To overcome this dilemma, we propose SMoA, a \textbf{S}pectrum \textbf{Mo}dulation \textbf{A}dapter that enlarges the accessible family of spectrum-aware updates under a smaller parameter budget. SMoA partitions the layer into multiple aligned spectral blocks and applies one in-block Hadamard-modulated low-rank branch to each diagonal block, yielding broader coverage of pretrained spectral directions. We provide theoretical analysis and empirical results on multiple tasks. In our experiments, SMoA improves average performance in the current lower-budget setting over LoRA and competitive LoRA-style baselines.

cs.LG↗
arXiv2026

ChunkFT: Byte-Streamed Optimization for Memory-Efficient Full Fine-Tuning

This work presents \textsc{ChunkFT}, a memory-efficient fine-tuning framework that reformulates full-parameter fine-tuning around a dynamically activated working set. \textsc{ChunkFT} enables gradient computation for arbitrary sub-tensors without modifying the network architecture, providing an algorithmic foundation for optimizing arbitrary sub-networks while avoiding standard dense gradient computation. We provide a theoretical convergence analysis of \textsc{ChunkFT} in the deterministic setting. Empirically, we apply \textsc{ChunkFT} to fine-tune Llama 3-8B and Llama 3-70B using a single RTX 4090-24GB GPU and 2$\times$ H800-80GB GPUs, respectively. Full-parameter fine-tuning of a 7B model with a 1K input length requires only 13.72GB of GPU memory. The results demonstrate the effectiveness of \textsc{ChunkFT} in memory usage, running time, and optimization quality. Moreover, downstream evaluations on language understanding, mathematical reasoning, and MT-Bench show that \textsc{ChunkFT} consistently outperforms existing memory-efficient baselines. Notably, \textsc{ChunkFT} achieves performance comparable to, and in some cases exceeding, full-parameter fine-tuning. Our repository is on https://github.com/misonsky/chunk.

cs.LG↗
arXiv2026

DiM\textsuperscript{3}: Bridging Multilingual and Multimodal Models via Direction- and Magnitude-Aware Merging

Towards more general and human-like intelligence, large language models should seamlessly integrate both multilingual and multimodal capabilities; however, extending an existing multimodal model to many languages typically requires expensive multilingual multimodal data construction and repeated end-to-end retraining. We study a training-free alternative: injecting multilingual capability into an existing multimodal model by composing residual updates in the shared language model backbone. The key challenge is that multilingual and multimodal updates are heterogeneous, reflecting different functional roles in the shared model. To address this, we propose Direction- and Magnitude-aware Multilingual Multimodal merging (DiM3), which selectively composes the two updates at each parameter dimension while preserving the original vision encoder and multimodal projector. Experiments on multilingual benchmarks in both text-only and vision-language settings, covering 57 languages across LLaVA- and Qwen-based backbones, show that DiM3 consistently outperforms existing merging baselines, substantially improves multilingual performance over the original multimodal model, and remains competitive with dedicated multilingual multimodal fine-tuning while largely retaining general multimodal ability. We further show that DiM3 can be directly applied to already trained multilingual multimodal models and still yield additional gains. Further interpretability analysis shows that DiM3 primarily reshapes intermediate-layer semantic representations, strengthening cross-lingual alignment under both text-only and multimodal inputs while preserving higher-layer task-sensitive structure. Our repository is on https://github.com/wzj1718/DiM3.

cs.CL↗
arXiv2026

SR-Nav: Spatial Relationships Matter for Zero-shot Object Goal Navigation

Zero-shot object-goal navigation aims to find target objects in unseen environments using only egocentric observation. Recent methods leverage foundation models' comprehension and reasoning capabilities to enhance navigation performance. However, when faced with poor viewpoints or weak semantic cues, foundation models often fail to support reliable reasoning in both perception and planning, resulting in inefficient or failed navigation. We observe that inherent relationships among objects and regions encode structured scene priors, which help agents infer plausible target locations even under partial observations. Motivated by this insight, we propose Spatial Relation-aware Navigation (SR-Nav), a framework that models both observed and experience-based spatial relationships to enhance both perception and planning. Specifically, SR-Nav first constructs a Dynamic Spatial Relationship Graph (DSRG) that encodes the target-centered spatial relationships through the foundation models and updates dynamically with real-time observations. We then introduce a Relation-aware Matching Module. It utilizes relationship matching instead of naive detection, leveraging diverse relationships in the DSRG to verify and correct errors, enhancing visual perception robustness. Finally, we design a Dynamic Relationship Planning Module to reduce the planning search space by dynamically computing the optimal paths based on the DSRG from the current position, thereby guiding planning and reducing exploration redundancy. Experiments on HM3D show that our method achieves state-of-the-art performance in both success rate and navigation efficiency. The code will be publicly available at https://github.com/Mzyw-1314/SR-Nav

cs.CV↗
arXiv2026

Resilient and Freshness-Aware Scheduling for Industrial Multi-Hop IAB Networks: A Packet Duplication Approach

In industrial millimeter-wave (mmWave) multi-hop Integrated Access and Backhaul (IAB) networks, dynamic blockages caused by moving obstacles pose a severe threat to robust and continuous networks. While Packet Duplication (PD) enhances reliability by path diversity, it inevitably doubles the traffic load, leading to severe congestion and degraded Age of Information (AoI). To navigate this reliability-congestion trade-off, we formulated an optimization problem in a multi-hop IAB scenario that minimizes the average AOI while satisfying strict queue stability constraints. We utilize Lyapunov optimization to transform the long-term stochastic optimization problem into tractable deterministic sub-problems. To solve these sub-problems efficiently, we propose a Resilient and Freshness-Aware Scheduling (RFAS) algorithm. Simulation results show that in blockage-prone environments, RFAS significantly outperforms baselines by maintaining a Packet Delivery Ratio (PDR) above 95\%. Crucially, it strictly guarantees queue stability under hard buffer constraints, whereas baselines suffer from buffer overflows. Furthermore, RFAS reduces the network load imbalance by 19\% compared to the baseline in high-frequency traffic scenarios. This confirms RFAS as a robust and sustainable solution for real-time industrial control loops.

cs.NI↗
arXiv2026

High-Rank Structured Modulation for Parameter-Efficient Fine-Tuning

As the number of model parameters increases, parameter-efficient fine-tuning (PEFT) has become the go-to choice for tailoring pre-trained large language models. Low-rank Adaptation (LoRA) uses a low-rank update method to simulate full parameter fine-tuning, which is widely used to reduce resource requirements. However, decreasing the rank encounters challenges with limited representational capacity when compared to full parameter fine-tuning. We present \textbf{SMoA}, a high-rank \textbf{S}tructured \textbf{MO}dulation \textbf{A}dapter that uses fewer trainable parameters while maintaining a higher rank, thereby improving the model's representational capacity and offering improved performance potential. The core idea is to freeze the original pretrained weights and selectively amplify or suppress important features of the original weights across multiple subspaces. The subspace mechanism provides an efficient way to increase the capacity and complexity of a model. We conduct both theoretical analyses and empirical studies on various tasks. Experiment results show that SMoA outperforms LoRA and its variants on 10 tasks, with extensive ablation studies validating its effectiveness.

cs.CL↗
arXiv2026

PlaM: Training-Free Plateau-Guided Model Merging for Better Visual Grounding in MLLMs

Multimodal Large Language Models (MLLMs) rely on strong linguistic reasoning inherited from their base language models. However, multimodal instruction fine-tuning paradoxically degrades this text's reasoning capability, undermining multimodal performance. To address this issue, we propose a training-free framework to mitigate this degradation. Through layer-wise vision token masking, we reveal a common three-stage pattern in multimodal large language models: early-modal separation, mid-modal alignment, and late-modal degradation. By analyzing the behavior of MLLMs at different stages, we propose a plateau-guided model merging method that selectively injects base language model parameters into MLLMs. Experimental results based on five MLLMs on nine benchmarks demonstrate the effectiveness of our method. Attention-based analysis further reveals that merging shifts attention from diffuse, scattered patterns to focused localization on task-relevant visual regions. Our repository is on https://github.com/wzj1718/PlaM.

cs.CL↗
arXiv2025

Urban Air Mobility: A Review of Recent Advances in Communication, Management, and Sustainability

Urban Air Mobility (UAM) offers a transformative approach to addressing urban congestion, improving accessibility, and advancing environmental sustainability. Rapid progress has emerged in three tightly linked domains since 2020: (1) Communication, where dynamic spectrum allocation and low-altitude channel characterization support reliable air-ground data exchange; (2) UAM management, with novel air-traffic control concepts for dense, largely autonomous urban airspace; and (3) Sustainability, driven by energy-efficient propulsion, integrated charging infrastructure, and holistic environmental assessment. This paper reviews and synthesizes the latest research across these areas, compares the state-of-the-art solutions, and outlines the technological and infrastructural milestones that are critical to realizing a scalable, sustainable UAM ecosystem.

eess.SY↗
arXiv2025

Timeliness-Aware Joint Source and Channel Coding for Adaptive Image Transmission

Accurate and timely image transmission is critical for emerging time-sensitive applications such as remote sensing in satellite-assisted Internet of Things. However, the bandwidth limitation poses a significant challenge in existing wireless systems, making it difficult to fulfill the requirements of both high-fidelity and low-latency image transmission. Semantic communication is expected to break through the performance bottleneck by focusing on the transmission of goal-oriented semantic information rather than raw data. In this paper, we employ a new timeliness metric named the value of information (VoI) and propose an adaptive joint source and channel coding (JSCC) method for image transmission that simultaneously considers both reconstruction quality and timeliness. Specifically, we first design a JSCC framework for image transmission with adaptive code length. Next, we formulate a VoI maximization problem by optimizing the transmission code length of the adaptive JSCC under the reconstruction quality constraint. Then, a deep reinforcement learning-based algorithm is proposed to solve the optimization problem efficiently. Experimental results show that the proposed method significantly outperforms baseline schemes in terms of reconstruction quality and timeliness, particularly in low signal-to-noise ratio conditions, offering a promising solution for efficient and robust image transmission in time-sensitive wireless networks.

eess.SP↗
arXiv2025

Equivariant Localization of $K$-homological Euler Class for almost connected Lie Groups

Using the Witten deformation and localization algebra techniques, we compute the $G$-equivariant $K$-homology class of the de Rham operator on a proper cocompact $G$-spin manifold, where $G$ is an almost connected Lie group. By applying a $G$-invariant Morse-Bott perturbation, this class is localized near the zero set of the perturbation and can be identified explicitly with an element in the representation rings associated to some isotropy subgroups. The result yields an equivariant Poincar\'e-Hopf formula and supplies concise tools for equivariant index computations.

math.OA↗
arXiv2025

A Bi-Orthogonal Structure-Preserving eigensolver for large-scale linear response eigenvalue problem

The linear response eigenvalue problem, which arises from many scientific and engineering fields, is quite challenging numerically for large-scale sparse/dense system, especially when it has zero eigenvalues. Based on a direct sum decomposition of biorthogonal invariant subspaces and the minimization principles in the biorthogonal complement, using the structure of generalized nullspace, we propose a Bi-Orthogonal Structure-Preserving subspace iterative solver, which is stable, efficient, and of excellent parallel scalability. The biorthogonality is of essential importance and created by a modified Gram-Schmidt biorthogonalization (MGS-Biorth) algorithm. We naturally deflate out converged eigenvectors by computing the rest eigenpairs in the biorthogonal complementary subspace without introducing any artificial parameters. When the number of requested eigenpairs is large, we propose a moving mechanism to compute them batch by batch such that the projection matrix size is small and independent of the requested eigenpair number. For large-scale problems, one only needs to provide the matrix-vector product, thus waiving explicit matrix storage. The numerical performance is further improved when the matrix-vector product is implemented using parallel computing. Ample numerical examples are provided to demonstrate the stability, efficiency, and parallel scalability.

math.NA↗
arXiv2025

Look Within or Look Beyond? A Theoretical Comparison Between Parameter-Efficient and Full Fine-Tuning

Parameter-Efficient Fine-Tuning (PEFT) methods achieve performance comparable to Full Fine-Tuning (FFT) while requiring significantly fewer computing resources, making it the go-to choice for researchers. We find that although PEFT can achieve competitive results on some benchmarks, its performance falls short of FFT in complex tasks, such as reasoning and instruction-based fine-tuning. In this paper, we compare the characteristics of PEFT and FFT in terms of representational capacity and robustness based on optimization theory. We theoretically demonstrate that PEFT is a strict subset of FFT. By providing theoretical upper bounds for PEFT, we show that the limited parameter space constrains the model's representational ability, making it more susceptible to perturbations. Experiments on 15 datasets encompassing classification, generation, reasoning, instruction fine-tuning tasks and 11 adversarial test sets validate our theories. We hope that these results spark further research beyond the realms of well established PEFT. The source code is in the anonymous Github repository\footnote{https://github.com/misonsky/PEFTEval}.

cs.LG↗
arXiv2025

Why Do More Experts Fail? A Theoretical Analysis of Model Merging

Model merging dramatically reduces storage and computational resources by combining multiple expert models into a single multi-task model. Although recent model merging methods have shown promising results, they struggle to maintain performance gains as the number of merged models increases. In this paper, we investigate the key obstacles that limit the scalability of model merging when integrating a large number of expert models. First, we prove that there is an upper bound on model merging. Further theoretical analysis reveals that the limited effective parameter space imposes a strict constraint on the number of models that can be successfully merged. Gaussian Width shows that the marginal benefit of merging additional models diminishes according to a strictly concave function. This implies that the effective parameter space becomes rapidly saturated as the number of merged models increases. Furthermore, using Approximate Kinematics Theory, we prove the existence of a unique optimal threshold beyond which adding more models does not yield significant performance improvements. At the same time, we introduce a straightforward Reparameterized Heavy-Tailed method (RHT) to extend the coverage of the merged model, thereby enhancing its performance. Empirical results on 12 benchmarks, including both knowledge-intensive and general-purpose tasks, validate our theoretical analysis. We believe that these results spark further research beyond the current scope of model merging. The source code is in the Github repository: https://github.com/wzj1718/ModelMergingAnalysis.

cs.LG↗
arXiv2025

PASE: A Massively Parallel Augmented Subspace Eigensolver for Large Scale Eigenvalue Problems

In this paper, we present a novel parallel augmented subspace method and build a package Parallel Augmented Subspace Eigensolver (PASE) for solving large scale eigenvalue problems by the massively parallel finite element discretization. Based on the augmented subspace, solving high dimensional eigenvalue problems can be transformed to solving the corresponding linear equations and low dimensional eigenvalue problems on the augmented subspace. Thus the complexity of solving the eigenvalue problems by augmented subspace method will be comparable to that of solving the same dimensinal linear equations. In order to improve the scalability and efficiency, we also present some implementing techniques for the parallel augmented subspace method. Based on parallel augmented subspace method and the concerned implementing techniques, a package PASE is built for solving large scale eigenvalue problems. Some numerical examples are provided to validate the efficiency and scalability of the proposed numerical methods.

math.NA↗