arXiv subjects

Xiao Zhang

Publications and source records attributed to Xiao Zhang.

At least 127 records · Page 7Linked to original sources

arXiv2025

Distributed Bilevel Optimization with Dual Pruning for Resource-limited Clients

With the development of large-scale models, traditional distributed bilevel optimization algorithms cannot be applied directly in low-resource clients. The key reason lies in the excessive computation involved in optimizing both the lower- and upper-level functions. Thus, we present the first resource-adaptive distributed bilevel optimization framework with a second-order free hypergradient estimator, which allows each client to optimize the submodels adapted to the available resources. Due to the coupled influence of partial outer parameters x and inner parameters y, it's challenging to theoretically analyze the upper bound regarding the globally averaged hypergradient for full model parameters. The error bound of inner parameter also needs to be reformulated since the local partial training. The provable theorems show that both RABO and RAFBO can achieve an asymptotically optimal convergence rate of $O(1/\sqrt{C_x^{\ast}Q})$, which is dominated by the minimum coverage of the outer parameter $C_x^{\ast}$. Extensive experiments on two different tasks demonstrate the effectiveness and computation efficiency of our proposed methods.

cs.DC↗

arXiv2025

Jailbreaking Attacks vs. Content Safety Filters: How Far Are We in the LLM Safety Arms Race?

As large language models (LLMs) are increasingly deployed, ensuring their safe use is paramount. Jailbreaking, adversarial prompts that bypass model alignment to trigger harmful outputs, present significant risks, with existing studies reporting high success rates in evading common LLMs. However, previous evaluations have focused solely on the models, neglecting the full deployment pipeline, which typically incorporates additional safety mechanisms like content moderation filters. To address this gap, we present the first systematic evaluation of jailbreak attacks targeting LLM safety alignment, assessing their success across the full inference pipeline, including both input and output filtering stages. Our findings yield two key insights: first, nearly all evaluated jailbreak techniques can be detected by at least one safety filter, suggesting that prior assessments may have overestimated the practical success of these attacks; second, while safety filters are effective in detection, there remains room to better balance recall and precision to further optimize protection and user experience. We highlight critical gaps and call for further refinement of detection accuracy and usability in LLM safety systems.

cs.CR↗

arXiv2025

Entropy-Guided Token Dropout: Training Autoregressive Language Models with Limited Domain Data

As access to high-quality, domain-specific data grows increasingly scarce, multi-epoch training has become a practical strategy for adapting large language models (LLMs). However, autoregressive models often suffer from performance degradation under repeated data exposure, where overfitting leads to a marked decline in model capability. Through empirical analysis, we trace this degradation to an imbalance in learning dynamics: predictable, low-entropy tokens are learned quickly and come to dominate optimization, while the model's ability to generalize on high-entropy tokens deteriorates with continued training. To address this, we introduce EntroDrop, an entropy-guided token dropout method that functions as structured data regularization. EntroDrop selectively masks low-entropy tokens during training and employs a curriculum schedule to adjust regularization strength in alignment with training progress. Experiments across model scales from 0.6B to 8B parameters show that EntroDrop consistently outperforms standard regularization baselines and maintains robust performance throughout extended multi-epoch training. These findings underscore the importance of aligning regularization with token-level learning dynamics when training on limited data. Our approach offers a promising pathway toward more effective adaptation of LLMs in data-constrained domains.

cs.CL↗

arXiv2025

Nearly-room-temperature ferromagnetism and tunable anomalous Hall effect in atomically thin Fe4CoGeTe2

Itinerant ferromagnetism at room temperature is a key ingredient for spin transport and manipulation. Here, we report the realization of nearly-room-temperature itinerant ferromagnetism in Co doped Fe5GeTe2 thin flakes. The ferromagnetic transition temperature TC (~ 323 K - 337 K) is almost unchanged when thickness is down to 12 nm and is still about 284 K at 2 nm (bilayer thickness). Theoretical calculations further indicate that the ferromagnetism persists in monolayer Fe4CoGeTe2. In addition to the robust ferromagnetism down to the ultrathin limit, Fe4CoGeTe2 exhibits an unusual temperature- and thickness-dependent intrinsic anomalous Hall effect. We propose that it could be ascribed to the dependence of band structure on thickness that changes the Berry curvature near the Fermi energy level subtly. The nearly-room-temperature ferromagnetism and tunable anomalous Hall effect in atomically thin Fe4CoGeTe2 provide opportunities to understand the exotic transport properties of two-dimensional van der Waals magnetic materials and explore their potential applications in spintronics.

cond-mat.mtrl-sci↗

arXiv2025

Data-Free Continual Learning of Server Models in Model-Heterogeneous Cloud-Device Collaboration

The rise of cloud-device collaborative computing has enabled intelligent services to be delivered across distributed edge devices while leveraging centralized cloud resources. In this paradigm, federated learning (FL) has become a key enabler for privacy-preserving model training without transferring raw data from edge devices to the cloud. However, with the continuous emergence of new data and increasing model diversity, traditional federated learning faces significant challenges, including inherent issues of data heterogeneity, model heterogeneity and catastrophic forgetting, along with new challenge of knowledge misalignment. In this study, we introduce FedDCL, a novel framework designed to enable data-free continual learning of the server model in a model-heterogeneous federated setting. We leverage pre-trained diffusion models to extract lightweight class-specific prototypes, which confer a threefold data-free advantage, enabling: (1) generation of synthetic data for the current task to augment training and counteract non-IID data distributions; (2) exemplar-free generative replay for retaining knowledge from previous tasks; and (3) data-free dynamic knowledge transfer from heterogeneous devices to the cloud server.Experimental results on various datasets demonstrate the effectiveness of FedDCL, showcasing its potential to enhance the generalizability and practical applicability of federated cloud-device collaboration in dynamic settings.

cs.LG↗

arXiv2025

Do MLLMs Really Understand the Charts?

Although Multimodal Large Language Models (MLLMs) have demonstrated increasingly impressive performance in chart understanding, most of them exhibit alarming hallucinations and significant performance degradation when handling non-annotated charts. We argue that current MLLMs rely largely on visual recognition rather than visual reasoning to interpret the charts, and visual estimation of numerical values is one of the most fundamental capabilities in chart understanding that require complex visual reasoning. To prove this, we introduce ChartVRBench, a benchmark meticulously designed to isolate and evaluate visual reasoning ability in chart understanding. Furthermore, we propose ChartVR-3B/7B trained with a novel Visual Reasoning Reinforcement Finetuning (VR-RFT) strategy to strengthen genuine chart visual reasoning abilities. Extensive experiments show that ChartVR achieves superior performance on ChartVRBench, outperforming even powerful proprietary models. Moreover, the visual reasoning skills cultivated by the proposed VR-RFT demonstrate strong generalization, leading to significant performance gains across a diverse suite of public chart understanding benchmarks. The code and dataset will be publicly available upon publication.

cs.CL↗

arXiv2025

Instance-Adaptive Keypoint Learning with Local-to-Global Geometric Aggregation for Category-Level Object Pose Estimation

Category-level object pose estimation aims to predict the 6D pose and size of previously unseen instances from predefined categories, requiring strong generalization across diverse object instances. Although many previous methods attempt to mitigate intra-class variations, they often struggle with instances exhibiting complex geometries or significant deviations from canonical shapes. To address this issue, we propose INKL-Pose, a novel category-level object pose estimation framework that enables INstance-adaptive Keypoint Learning with local-to-global geometric aggregation. Specifically, our method first predicts semantically consistent and geometrically informative keypoints using an Instance-Adaptive Keypoint Detector, then refines them: (1) a Local Keypoint Feature Aggregator capturing fine-grained geometries, and (2) a Global Keypoint Feature Aggregator using bidirectional Mamba for structural consistency. To enable bidirectional modeling in Mamba, we introduce a simple yet effective Feature Sequence Flipping strategy that preserves spatial coherence while constructing backward feature sequence. Additionally, we design a surface loss and a separation loss to encourage uniform coverage and spatial diversity in keypoint distribution. The resulting keypoints are mapped to a canonical space for 6D pose and size regression. Extensive experiments on CAMERA25, REAL275, and HouseCat6D show that INKL-Pose achieves state-of-the-art performance with 16.7M parameters and runs at 36 FPS on an NVIDIA RTX 4090D GPU.

cs.CV↗

arXiv2025

Massless Majorana spinors in the Kerr spacetime

In this paper, we show that massive Majorana spinors \eqref{1.2} do not exist if they are $t$-dependent or $ϕ$-dependent in Kerr, or Kerr-(A)dS spacetimes. For massless Majorana spinors in the non-extreme Kerr spacetime, the Dirac equation can be separated into radial and angular equations, parameterized by two complex constants $ε_1$, $ε_2$. If at least one of $ε_1$, $ε_2$ is zero, massless Majorana spinors can be solved explicitly. If $ε_1$, $ε_2$ are nonzero, we prove the nonexistence of massless time-periodic Majorana spinors in the non-extreme Kerr spacetime which are $L^p$ outside the event horizon for $ 0<p\le\frac{6}{|ε_1|+|ε_2| +2}$. We then provide the Hamiltonian formulation for massless Majorana spinors and prove that the self-adjointness of the Hamiltonian leads to the angular momentum $a=0$ and spacetime reduces to the Schwarzschild spacetime, moreover, the massless Majorana spinor must be $ϕ$-independent. Finally, we show that, in the Schwarzschild spacetime, for initial data with $L^2$ decay at infinity, the probability of the massless Majorana spinors to be in any compact region of space tends to zero as time tends to infinity.

gr-qc↗

arXiv2025

Revisiting Clustering of Neural Bandits: Selective Reinitialization for Mitigating Loss of Plasticity

Clustering of Bandits (CB) methods enhance sequential decision-making by grouping bandits into clusters based on similarity and incorporating cluster-level contextual information, demonstrating effectiveness and adaptability in applications like personalized streaming recommendations. However, when extending CB algorithms to their neural version (commonly referred to as Clustering of Neural Bandits, or CNB), they suffer from loss of plasticity, where neural network parameters become rigid and less adaptable over time, limiting their ability to adapt to non-stationary environments (e.g., dynamic user preferences in recommendation). To address this challenge, we propose Selective Reinitialization (SeRe), a novel bandit learning framework that dynamically preserves the adaptability of CNB algorithms in evolving environments. SeRe leverages a contribution utility metric to identify and selectively reset underutilized units, mitigating loss of plasticity while maintaining stable knowledge retention. Furthermore, when combining SeRe with CNB algorithms, the adaptive change detection mechanism adjusts the reinitialization frequency according to the degree of non-stationarity, ensuring effective adaptation without unnecessary resets. Theoretically, we prove that SeRe enables sublinear cumulative regret in piecewise-stationary environments, outperforming traditional CNB approaches in long-term performances. Extensive experiments on six real-world recommendation datasets demonstrate that SeRe-enhanced CNB algorithms can effectively mitigate the loss of plasticity with lower regrets, improving adaptability and robustness in dynamic settings.

cs.LG↗

arXiv2025

VALUE: Value-Aware Large Language Model for Query Rewriting via Weighted Trie in Sponsored Search

Query-to-bidword(i.e., bidding keyword) rewriting is fundamental to sponsored search, transforming noisy user queries into semantically relevant and commercially valuable keywords. Recent advances in large language models (LLMs) improve semantic relevance through generative retrieval frameworks, but they rarely encode the commercial value of keywords. As a result, rewrites are often semantically correct yet economically suboptimal, and a reinforcement learning from human feedback (RLHF) stage is usually added after supervised fine-tuning(SFT) to mitigate this deficiency. However, conventional preference alignment frequently overemphasize the ordering of bidword values and is susceptible to overfitting, which degrades rewrite quality. In addition, bidword value changes rapidly, while existing generative methods do not respond to these fluctuations. To address this shortcoming, we introduce VALUE(Value-Aware Large language model for qUery rewriting via wEighted trie), a framework that integrates value awareness directly into generation and enhances value alignment during training. VALUE employs the Weighted Trie, a novel variant of the classical trie that stores real-time value signals for each token. During decoding, the framework adjusts the LLM's token probabilities with these signals, constraining the search space and steering generation toward high-value rewrites. The alignment stage uses a fine-grained preference learning strategy that emphasizes stable, high-value differences and down-weights noisy or transient fluctuations, thereby improving robustness and reducing overfitting. Offline experiments show that VALUE significantly outperforms baselines in both semantic matching and value-centric metrics. VALUE has been deployed on our advertising system since October 2024 and served the Double Eleven promotions, the biggest shopping carnival in China.

cs.IR↗

arXiv2025

VividFace: High-Quality and Efficient One-Step Diffusion For Video Face Enhancement

Video Face Enhancement (VFE) aims to restore high-quality facial regions from degraded video sequences, enabling a wide range of practical applications. Despite substantial progress in the field, current methods that primarily rely on video super-resolution and generative frameworks continue to face three fundamental challenges: (1) computational inefficiency caused by iterative multi-step denoising in diffusion models; (2) faithfully modeling intricate facial textures while preserving temporal consistency; and (3) limited model generalization due to the lack of high-quality face video training data. To address these challenges, we propose VividFace, a novel and efficient one-step diffusion framework for VFE. Built upon the pretrained WANX video generation model, VividFace reformulates the traditional multi-step diffusion process as a single-step flow matching paradigm that directly maps degraded inputs to high-quality outputs with significantly reduced inference time. To enhance facial detail recovery, we introduce a Joint Latent-Pixel Face-Focused Training strategy that constructs spatiotemporally aligned facial masks to guide optimization toward critical facial regions in both latent and pixel spaces. Furthermore, we develop an MLLM-driven automated filtering pipeline that produces MLLM-Face90, a meticulously curated high-quality face video dataset, ensuring models learn from photorealistic facial textures. Extensive experiments demonstrate that VividFace achieves superior performance in perceptual quality, identity preservation, and temporal consistency across both synthetic and real-world benchmarks. We will publicly release our code, models, and dataset to support future research.

cs.CV↗

arXiv2025

Preference-Guided Reinforcement Learning for Efficient Exploration

In this paper, we investigate preference-based reinforcement learning (PbRL), which enables reinforcement learning (RL) agents to learn from human feedback. This is particularly valuable when defining a fine-grain reward function is not feasible. However, this approach is inefficient and impractical for promoting deep exploration in hard-exploration tasks with long horizons and sparse rewards. To tackle this issue, we introduce LOPE: \textbf{L}earning \textbf{O}nline with trajectory \textbf{P}reference guidanc\textbf{E}, an end-to-end preference-guided RL framework that enhances exploration efficiency in hard-exploration tasks. Our intuition is that LOPE directly adjusts the focus of online exploration by considering human feedback as guidance, thereby avoiding the need to learn a separate reward model from preferences. Specifically, LOPE includes a two-step sequential policy optimization technique consisting of trust-region-based policy improvement and preference guidance steps. We reformulate preference guidance as a trajectory-wise state marginal matching problem that minimizes the maximum mean discrepancy distance between the preferred trajectories and the learned policy. Furthermore, we provide a theoretical analysis to characterize the performance improvement bound and evaluate the effectiveness of the LOPE. When assessed in various challenging hard-exploration environments, LOPE outperforms several state-of-the-art methods in terms of convergence rate and overall performance.The code used in this study is available at https://github.com/buaawgj/LOPE.

cs.LG↗

arXiv2025

LLaDA-Rec: Discrete Diffusion for Parallel Semantic ID Generation in Generative Recommendation

Generative recommendation represents each item as a semantic ID, i.e., a sequence of discrete tokens, and generates the next item through autoregressive decoding. While effective, existing autoregressive models face two intrinsic limitations: (1) unidirectional constraints, where causal attention restricts each token to attend only to its predecessors, hindering global semantic modeling; and (2) error accumulation, where the fixed left-to-right generation order causes prediction errors in early tokens to propagate to the predictions of subsequent token. To address these issues, we propose LLaDA-Rec, a discrete diffusion framework that reformulates recommendation as parallel semantic ID generation. By combining bidirectional attention with the adaptive generation order, the approach models inter-item and intra-item dependencies more effectively and alleviates error accumulation. Specifically, our approach comprises three key designs: (1) a parallel tokenization scheme that produces semantic IDs for bidirectional modeling, addressing the mismatch between residual quantization and bidirectional architectures; (2) two masking mechanisms at the user-history and next-item levels to capture both inter-item sequential dependencies and intra-item semantic relationships; and (3) an adapted beam search strategy for adaptive-order discrete diffusion decoding, resolving the incompatibility of standard beam search with diffusion-based generation. Experiments on three real-world datasets show that LLaDA-Rec consistently outperforms both ID-based and state-of-the-art generative recommenders, establishing discrete diffusion as a new paradigm for generative recommendation.

cs.IR↗

arXiv2025

GASP: Efficient Black-Box Generation of Adversarial Suffixes for Jailbreaking LLMs

LLMs have shown impressive capabilities across various natural language processing tasks, yet remain vulnerable to input prompts, known as jailbreak attacks, carefully designed to bypass safety guardrails and elicit harmful responses. Traditional methods rely on manual heuristics but suffer from limited generalizability. Despite being automatic, optimization-based attacks often produce unnatural prompts that can be easily detected by safety filters or require high computational costs due to discrete token optimization. In this paper, we introduce Generative Adversarial Suffix Prompter (GASP), a novel automated framework that can efficiently generate human-readable jailbreak prompts in a fully black-box setting. In particular, GASP leverages latent Bayesian optimization to craft adversarial suffixes by efficiently exploring continuous latent embedding spaces, gradually optimizing the suffix prompter to improve attack efficacy while balancing prompt coherence via a targeted iterative refinement procedure. Through comprehensive experiments, we show that GASP can produce natural adversarial prompts, significantly improving jailbreak success over baselines, reducing training times, and accelerating inference speed, thus making it an efficient and scalable solution for red-teaming LLMs.

cs.LG↗

arXiv2025

TAMO: Fine-Grained Root Cause Analysis via Tool-Assisted LLM Agent with Multi-Modality Observation Data in Cloud-Native Systems

Implementing large language models (LLMs)-driven root cause analysis (RCA) in cloud-native systems has become a key topic of modern software operations and maintenance. However, existing LLM-based approaches face three key challenges: multi-modality input constraint, context window limitation, and dynamic dependence graph. To address these issues, we propose a tool-assisted LLM agent with multi-modality observation data for fine-grained RCA, namely TAMO, including multimodality alignment tool, root cause localization tool, and fault types classification tool. In detail, TAMO unifies multi-modal observation data into time-aligned representations for cross-modal feature consistency. Based on the unified representations, TAMO then invokes its specialized root cause localization tool and fault types classification tool for further identifying root cause and fault type underlying system context. This approach overcomes the limitations of LLMs in processing real-time raw observational data and dynamic service dependencies, guiding the model to generate repair strategies that align with system context through structured prompt design. Experiments on two benchmark datasets demonstrate that TAMO outperforms state-of-the-art (SOTA) approaches with comparable performance.

cs.AI↗

arXiv2025

PROFIT: A Specialized Optimizer for Deep Fine Tuning

The fine-tuning of pre-trained models has become ubiquitous in generative AI, computer vision, and robotics. Although much attention has been paid to improving the efficiency of fine-tuning model, there has been less scholarship around fine-tuning specifically for improved model performance. To remedy this gap, we present PROFIT, one of the first optimizers designed to incrementally fine-tune converged models on new tasks and/or datasets. Unlike traditional optimizers such as SGD or Adam, which make minimal assumptions due to random initializations, PROFIT takes the properties of a converged model into account explicitly to regularize the optimization process. Employing a temporal gradient-orthogonalization process, PROFIT outperforms fine-tuning methods in various tasks, from image classification to multimodal language model training to large-scale motion prediction. Moreover, PROFIT is encapsulated as a modular optimizer, which makes it easy to integrate directly into any training pipeline with minimal engineering effort.

cs.CV↗

arXiv2025

Backmapping of the High- and Low-latitude Solar Wind under Multiple Heliospheric and Coronal Magnetic Field Configurations

Solar wind backmapping is a critical technique for analyzing the origin of the solar wind and space weather events by correlating in situ measurements with solar remote-sensing observations. This technique typically traces magnetic field lines using a heliospheric magnetic field (HMF) model coupled with a coronal magnetic field (CMF). However, the impact of different HMF and CMF configurations on backmapping uncentainty-particularly regarding high-latitude solar wind-remains inadequately quantified. This study comprehensively evaluates solar wind backmapping by combining two HMF models (Parker spiral, Fisk-type) with three CMF models (Potential Field Source Surface (PFSS), Potential Field Current Sheet (PFCS), Current Sheet Source Surface (CSSS)). Our analysis primarily uses in situ measurements from Ulysses and remote-sensing data from STEREO-A. Key findings are that: (1) while both Fisk and Parker HMF models show comparable consistency with measured magnetic field strength and polarity, they produce certain longitudinal displacements in their back-mapped footpoints on the source surface (2.5$R_{\odot}$); (2) For CMF models (PFSS, PFCS, CSSS), predicted photospheric footpoints exhibit minor variations for high/mid-latitude solar wind but some divergences for ecliptic/low-latitude wind; (3) All three CMF models link high/mid-latitude wind to active regions or coronal holes, yet associate a fraction of ecliptic/low-latitude wind with quiet-Sun regions; (4) Ecliptic/low-latitude sources show significantly stronger dependence on the PFSS source surface height compared to high-latitude wind. These results demonstrate that simpler models (PFSS + Parker) appear reasonably adequate for polar coronal hole wind studies, while low-latitude/ecliptic solar wind exhibits the heightened sensitivity to model choices.

astro-ph.SR↗

arXiv2025

Unity is Power: Semi-Asynchronous Collaborative Training of Large-Scale Models with Structured Pruning in Resource-Limited Clients

In this work, we study to release the potential of massive heterogeneous weak computing power to collaboratively train large-scale models on dispersed datasets. In order to improve both efficiency and accuracy in resource-adaptive collaborative learning, we take the first step to consider the \textit{unstructured pruning}, \textit{varying submodel architectures}, \textit{knowledge loss}, and \textit{straggler} challenges simultaneously. We propose a novel semi-asynchronous collaborative training framework, namely ${Co\text{-}S}^2{P}$, with data distribution-aware structured pruning and cross-block knowledge transfer mechanism to address the above concerns. Furthermore, we provide theoretical proof that ${Co\text{-}S}^2{P}$ can achieve asymptotic optimal convergence rate of $O(1/\sqrt{N^*EQ})$. Finally, we conduct extensive experiments on two types of tasks with a real-world hardware testbed including diverse IoT devices.The experimental results demonstrate that $Co\text{-}S^2P$ improves accuracy by up to 8.8\% and resource utilization by up to 1.2$\times$ compared to state-of-the-art methods, while reducing memory consumption by approximately 22\% and training time by about 24\% on all resource-limited devices.

cs.DC↗