arXiv subjects

Jie Zhang

Publications and source records attributed to Jie Zhang.

At least 343 records · Page 19Linked to original sources

arXiv2025

SafeGuider: Robust and Practical Content Safety Control for Text-to-Image Models

Text-to-image models have shown remarkable capabilities in generating high-quality images from natural language descriptions. However, these models are highly vulnerable to adversarial prompts, which can bypass safety measures and produce harmful content. Despite various defensive strategies, achieving robustness against attacks while maintaining practical utility in real-world applications remains a significant challenge. To address this issue, we first conduct an empirical study of the text encoder in the Stable Diffusion (SD) model, which is a widely used and representative text-to-image model. Our findings reveal that the [EOS] token acts as a semantic aggregator, exhibiting distinct distributional patterns between benign and adversarial prompts in its embedding space. Building on this insight, we introduce SafeGuider, a two-step framework designed for robust safety control without compromising generation quality. SafeGuider combines an embedding-level recognition model with a safety-aware feature erasure beam search algorithm. This integration enables the framework to maintain high-quality image generation for benign prompts while ensuring robust defense against both in-domain and out-of-domain attacks. SafeGuider demonstrates exceptional effectiveness in minimizing attack success rates, achieving a maximum rate of only 5.48\% across various attack scenarios. Moreover, instead of refusing to generate or producing black images for unsafe prompts, SafeGuider generates safe and meaningful images, enhancing its practical utility. In addition, SafeGuider is not limited to the SD model and can be effectively applied to other text-to-image models, such as the Flux model, demonstrating its versatility and adaptability across different architectures. We hope that SafeGuider can shed some light on the practical deployment of secure text-to-image systems.

cs.CR↗

arXiv2025

Oyster-I: Beyond Refusal -- Constructive Safety Alignment for Responsible Language Models

Large language models (LLMs) typically deploy safety mechanisms to prevent harmful content generation. Most current approaches focus narrowly on risks posed by malicious actors, often framing risks as adversarial events and relying on defensive refusals. However, in real-world settings, risks also come from non-malicious users seeking help while under psychological distress (e.g., self-harm intentions). In such cases, the model's response can strongly influence the user's next actions. Simple refusals may lead them to repeat, escalate, or move to unsafe platforms, creating worse outcomes. We introduce Constructive Safety Alignment (CSA), a human-centric paradigm that protects against malicious misuse while actively guiding vulnerable users toward safe and helpful results. Implemented in Oyster-I (Oy1), CSA combines game-theoretic anticipation of user reactions, fine-grained risk boundary discovery, and interpretable reasoning control, turning safety into a trust-building process. Oy1 achieves state-of-the-art safety among open models while retaining high general capabilities. On our Constructive Benchmark, it shows strong constructive engagement, close to GPT-5, and unmatched robustness on the Strata-Sword jailbreak dataset, nearing GPT-o1 levels. By shifting from refusal-first to guidance-first safety, CSA redefines the model-user relationship, aiming for systems that are not just safe, but meaningfully helpful. We release Oy1, code, and the benchmark to support responsible, user-centered AI.

cs.AI↗

arXiv2025

LouisKV: Efficient KV Cache Retrieval for Long Input-Output Sequences

While Key-Value (KV) cache succeeds in reducing redundant computations in auto-regressive models, it introduces significant memory overhead, limiting its practical deployment in long-sequence scenarios. Existing KV retrieval methods mitigate this by dynamically retaining only a subset of KV entries on the GPU. However, they still suffer from notable efficiency and accuracy bottlenecks due to per-token retrieval and coarse-grained page-level KV management, especially in long-output reasoning scenarios. With the emergence of large reasoning models, efficiently handling such scenarios has become increasingly important. To address this issue, we present two key observations: (1) critical KVs exhibit strong temporal locality during decoding, and (2) these KVs exhibit distinct distribution patterns across the input prompt and generated output. Building on these observations, we propose LouisKV, an efficient KV cache retrieval framework designed for various long-sequence scenarios. Specifically, LouisKV introduces a semantic-aware retrieval strategy leveraging temporal locality to trigger retrieval only at semantic boundaries, drastically reducing computation and data transfer overhead. LouisKV also designs a decoupled, fine-grained management scheme that tailors differentiated strategies for input and output sequences to create retrieval units that better match the model's attention patterns, enabling precise identification of critical KVs. Furthermore, to boost efficiency, LouisKV incorporates several kernel-level optimizations, including custom Triton and CUDA kernels to accelerate the KV clustering and retrieval. Evaluations show that LouisKV achieves up to 4.7$\times$ speedup over state-of-the-art KV retrieval methods while maintaining near-lossless accuracy across diverse long-sequence tasks, including long-input short-output, short-input long-output, and long-input long-output scenarios.

cs.LG↗

arXiv2025

PACEbench: A Framework for Evaluating Practical AI Cyber-Exploitation Capabilities

The increasing autonomy of Large Language Models (LLMs) necessitates a rigorous evaluation of their potential to aid in cyber offense. Existing benchmarks often lack real-world complexity and are thus unable to accurately assess LLMs' cybersecurity capabilities. To address this gap, we introduce PACEbench, a practical AI cyber-exploitation benchmark built on the principles of realistic vulnerability difficulty, environmental complexity, and cyber defenses. Specifically, PACEbench comprises four scenarios spanning single, blended, chained, and defense vulnerability exploitations. To handle these complex challenges, we propose PACEagent, a novel agent that emulates human penetration testers by supporting multi-phase reconnaissance, analysis, and exploitation. Extensive experiments with seven frontier LLMs demonstrate that current models struggle with complex cyber scenarios, and none can bypass defenses. These findings suggest that current models do not yet pose a generalized cyber offense threat. Nonetheless, our work provides a robust benchmark to guide the trustworthy development of future models.

cs.CR↗

arXiv2025

Magnetic Decay Index Profile and Coronal Mass Ejection Speed

We study the relationship between the speed of coronal mass ejections (CMEs) and the height profile of the ambient magnetic field, quantified by its decay index, n(h). Our sample is composed of 15 very fast CMEs (Vcme > 1500 km/s; all halo CMEs) and 22 halo CMEs below this speed limit from Solar Cycle 23. The very fast CMEs yield a high correlation of 0.81 between Vcme and the slope of n(h) in a height range above the onset height of the torus instability if one extremely fast outlier, which closely followed another very fast CME, is excluded. This is consistent with the hypothesis that the torus instability plays a decisive role in CME acceleration. The whole sample yields a weaker correlation, which is still significant if events with a broad torus-stable dip in n(h) are excluded. A parametric simulation study of flux-rope eruptions from quadrupolar and two-scale bipolar source regions confirms the decelerating effect of such dips. Very fast, moderate-velocity, and confined eruptions are found.

astro-ph.SR↗

arXiv2025

Optimal Planning of Electric Vehicle Charging Stations: Integrating Public Charging Networks and Transportation Congestion

The adoption of electric vehicles (EVs) represents a critical shift in personal mobility, fueled by policy support and advancements in automotive technology. However, the expansion of EVs for long-distance travel is hindered by charging time concerns, the sparse distribution of charging stations, and the worsening waiting times due to congestion. The main objective of this work is two-fold: 1) first, to comprehensively analyze the existing public charging station robustness and effectively strategize for the new ones, and 2) secondly, to select the optimal chargers for long-distance journeys, by estimating the waiting time from current traffic congestion. This is achieved by accompanying effective EV charging strategies, pinpointing on the congestion points from the existing traffic, and the robustness of the current charging station infrastructure. Utilizing a real-time transportation and charging station dataset in Texas, we identify optimal charger placement strategies to minimize travel time by examining the congestion and charging time trade-offs. Our findings suggest that maximizing the constant current phase during charging enhances efficiency, crucial for long-distance travel. On the contrary, we also explore the negative impact of congestion on travel times and we conclude that sometimes it might be beneficial to exceed the constant current phase to avoid the congested charging stations.

eess.SY↗

arXiv2025

CFVBench: A Comprehensive Video Benchmark for Fine-grained Multimodal Retrieval-Augmented Generation

Multimodal Retrieval-Augmented Generation (MRAG) enables Multimodal Large Language Models (MLLMs) to generate responses with external multimodal evidence, and numerous video-based MRAG benchmarks have been proposed to evaluate model capabilities across retrieval and generation stages. However, existing benchmarks remain limited in modality coverage and format diversity, often focusing on single- or limited-modality tasks, or coarse-grained scene understanding. To address these gaps, we introduce CFVBench, a large-scale, manually verified benchmark constructed from 599 publicly available videos, yielding 5,360 open-ended QA pairs. CFVBench spans high-density formats and domains such as chart-heavy reports, news broadcasts, and software tutorials, requiring models to retrieve and reason over long temporal video spans while maintaining fine-grained multimodal information. Using CFVBench, we systematically evaluate 7 retrieval methods and 14 widely-used MLLMs, revealing a critical bottleneck: current models (even GPT5 or Gemini) struggle to capture transient yet essential fine-grained multimodal details. To mitigate this, we propose Adaptive Visual Refinement (AVR), a simple yet effective framework that adaptively increases frame sampling density and selectively invokes external tools when necessary. Experiments show that AVR consistently enhances fine-grained multimodal comprehension and improves performance across all evaluated MLLMs

cs.CL↗

arXiv2025

OBJVanish: Physically Realizable Text-to-3D Adv. Generation of LiDAR-Invisible Objects

LiDAR-based 3D object detectors are fundamental to autonomous driving, where failing to detect objects poses severe safety risks. Developing effective 3D adversarial attacks is essential for thoroughly testing these detection systems and exposing their vulnerabilities before real-world deployment. However, existing adversarial attacks that add optimized perturbations to 3D points have two critical limitations: they rarely cause complete object disappearance and prove difficult to implement in physical environments. We introduce the text-to-3D adversarial generation method, a novel approach enabling physically realizable attacks that can generate 3D models of objects truly invisible to LiDAR detectors and be easily realized in the real world. Specifically, we present the first empirical study that systematically investigates the factors influencing detection vulnerability by manipulating the topology, connectivity, and intensity of individual pedestrian 3D models and combining pedestrians with multiple objects within the CARLA simulation environment. Building on the insights, we propose the physically-informed text-to-3D adversarial generation (Phy3DAdvGen) that systematically optimizes text prompts by iteratively refining verbs, objects, and poses to produce LiDAR-invisible pedestrians. To ensure physical realizability, we construct a comprehensive object pool containing 13 3D models of real objects and constrain Phy3DAdvGen to generate 3D objects based on combinations of objects in this set. Extensive experiments demonstrate that our approach can generate 3D pedestrians that evade six state-of-the-art (SOTA) LiDAR 3D detectors in both CARLA simulation and physical environments, thereby highlighting vulnerabilities in safety-critical applications.

cs.CV↗

arXiv2025

Think How to Think: Mitigating Overthinking with Autonomous Difficulty Cognition in Large Reasoning Models

Recent Large Reasoning Models (LRMs) excel at complex reasoning tasks but often suffer from overthinking, generating overly long and redundant reasoning trajectories. To explore its essence, our empirical analysis reveals that LRMs are primarily limited to recognizing task properties (i.e., difficulty levels) like humans before solving the problem, leading to a one-size-fits-all reasoning process. Inspired by this, a pressing and natural question emerges: Can we explicitly bootstrap such ability to alleviate overthinking in LRMs? In this paper, we propose Think-How-to-Think (TH2T), a novel two-stage fine-tuning strategy that progressively inspires LRMs' difficulty cognition and redundancy cognition of LRMs. Specifically, we first inject difficulty hypnosis into output prefixes to guide the model toward adaptive reasoning depth, trained on a hybrid dataset mixing short and long reasoning paths. Then, we incorporate redundancy hypnosis, which supervises the intermediate reasoning steps to identify and eliminate unnecessary reasoning patterns. Experiments on 7B/14B/32B models demonstrate that TH2T significantly reduces inference costs by over 70% on easy tasks and 40% on hard tasks while maintaining performance stability. The resulting outputs exhibit clear signs of difficulty-aware capabilities and reduced redundancy (e.g., reflection and looping).

cs.AI↗

arXiv2025

Multiplayer General Lotto game

In this paper, we investigate the multiplayer General Lotto game across multiple battlefields, a significant variant of the Colonel Blotto game. In this version, each player employs a probability distribution for resource allocation, ensuring that their expected expenditure does not exceed their budget. We first establish the existence of the Nash equilibrium in a general setting, where players' budgets are asymmetric and the values of the battlefields are heterogeneous and asymmetric among players. Next, we provide a detailed characterization of the Nash equilibrium for multiple players on a single battlefield. In this characterization, we observe that the upper endpoints of the supports of players' equilibrium strategies coincide, and that the minimum value of a player's support above zero inversely correlates with his budget. We demonstrate the uniqueness of Nash equilibrium over a single battlefield in some scenarios. In the multi-battlefield setting, we prove that there is an upper bound on the average number of battlefields each player participates in. Additionally, we provide an example demonstrating the non-uniqueness of the Nash equilibrium in the context of multiple battlefields with multiple players. Finally, we present a solution for the Nash equilibrium in a symmetric case.

cs.GT↗

arXiv2025

Knowledge-Aware Modeling with Frequency Adaptive Learning for Battery Health Prognostics

Battery health prognostics are critical for ensuring safety, efficiency, and sustainability in modern energy systems. However, it has been challenging to achieve accurate and robust prognostics due to complex battery degradation behaviors with nonlinearity, noise, capacity regeneration, etc. Existing data-driven models capture temporal degradation features but often lack knowledge guidance, which leads to unreliable long-term health prognostics. To overcome these limitations, we propose Karma, a knowledge-aware model with frequency-adaptive learning for battery capacity estimation and remaining useful life prediction. The model first performs signal decomposition to derive battery signals in different frequency bands. A dual-stream deep learning architecture is developed, where one stream captures long-term low-frequency degradation trends and the other models high-frequency short-term dynamics. Karma regulates the prognostics with knowledge, where battery degradation is modeled as a double exponential function based on empirical studies. Our dual-stream model is used to optimize the parameters of the knowledge with particle filters to ensure physically consistent and reliable prognostics and uncertainty quantification. Experimental study demonstrates Karma's superior performance, achieving average error reductions of 50.6% and 32.6% over state-of-the-art algorithms for battery health prediction on two mainstream datasets, respectively. These results highlight Karma's robustness, generalizability, and potential for safer and more reliable battery management across diverse applications.

cs.LG↗

arXiv2025

FinGAN: An Interpretable RSS Generation Network for Scalable Fingerprint Localization

This work introduces FinGAN, a robust received signal strength (RSS) data generator designed to expand RSS fingerprint datasets. Compared to existing generative adversarial models that either rely on known reference positions (RPs) or depend on predefined priors, FinGAN learns the latent information between RPs and RSS values by maximizing the mutual information between the generated RSS data and the RPs, enabling an end-to-end RSS generation directly from RPs. This allows us to accurately generate RSS data for previously unmeasured RPs. Both quantitative and qualitative evaluations demonstrate that FinGAN produces synthetic RSS data closely aligned with real RSS sample collected from the on-site experiment, preserving localization performance comparable to that achieved with complete real-world datasets. To further validate its generalizability, FinGAN is also trained and evaluated on open-source datasets from three typical office environments,and the results demonstrate consistent performance across different scenarios.

eess.SP↗

arXiv2025

Hybrid Powertrain Optimization for Regional Aircraft Integrating Hydrogen Fuel Cells and Aluminum Air Batteries

With the increasing demand for air travel and the urgency to reduce emissions, transitioning from fossil fuel-based propulsion systems is a critical step toward sustainable aviation. While batteries are widely used in urban air mobility, their long charging durations limit their feasibility for consecutive flights. Hybrid propulsion systems, which integrate fuel cells and batteries, offer a promising alternative due to their higher energy density and improved efficiency. This paper presents a novel hybrid powertrain architecture for regional aircraft, incorporating a hydrogen fuel cell, a lithium-ion battery, and an auxiliary aluminum-air battery. The proposed system is evaluated using real-world power demand data from a Cessna 208 aircraft. The hydrogen fuel cell acts as the primary power source, ensuring continuous operation, while the lithium-ion battery manages transient power fluctuations to enhance system stability. The aluminum-air battery is introduced as a high-energy emergency backup, providing extended endurance during critical situations. A mixed-integer optimization model is formulated for system sizing and power scheduling, ensuring optimal energy distribution among the power sources. Multiple operational scenarios are analyzed to evaluate system performance, particularly under emergency conditions, where power reliability is crucial. The results highlight the feasibility and effectiveness of the proposed hybrid architecture in improving energy efficiency and flight safety for regional aircraft applications.

math.OC↗

arXiv2025

Electric Vehicle Charger Infrastructure Planning: Demand Estimation, Coverage Optimization Over an Integrated Power Grid

For electrifying the transportation sector, deploying a strategically planned and efficient charging infrastructure is essential. This paper presents a two-phase approach for electric vehicle (EV) charger deployment that integrates spatial point-of-interest analysis and maximum coverage optimization over an integrated spatial power grid. Spatial-focused studies in the literature often overlook electrical grid constraints, while grid-focused work frequently considers statistically modeled EV charging demand. To address these gaps, a new framework is proposed that combines spatial network planning with electrical grid considerations. This study approaches EV charger planning from a perspective of the distribution grid, starting with an estimation of EV charging demand and the identification of optimal candidate locations. It ensures that the capacity limits of newly established chargers are maintained within the limits of the power grid. This framework is applied in a test case for the Dallas area, integrating the existing EV charger network with an 8500-bus distribution system for comprehensive planning.

eess.SY↗

arXiv2025

VeriFuzzy: A Dynamic Verifiable Fuzzy Search Service for Encrypted Cloud Data

Enabling search over encrypted cloud data is essential for privacy-preserving data outsourcing. While searchable encryption has evolved to support individual requirements like fuzzy matching, dynamic updates, and result verification, designing a service that supports dynamic, verifiable fuzzy search (DVFS) over encrypted cloud data remains a fundamental challenge due to inherent conflicts between underlying technologies. Existing approaches struggle with simultaneously achieving efficiency, functionality, and security, often forcing impractical trade-offs. This paper presents \textbf{VeriFuzzy}, a novel DVFS service framework that cohesively integrates three innovations: an \textit{Enhanced Virtual Binary Tree (EVBTree)} that decouples fuzzy semantics from index logic to support $O(\log n)$ search/updates; a \textit{blockchain-reconstructed verification} mechanism that ensures result integrity with logarithmic complexity; and a \textit{dual-repository state management} scheme that achieves IND-CKA2 security by neutralizing branch leakage. Extensive evaluation on 3,500+ documents shows VeriFuzzy achieves 41\% faster search, $5\times$ more efficient verification, and constant-time index updates compared to state-of-the-art alternatives. Our code and dataset are now open source, hoping to inspire future DVFS research.

cs.CR↗

arXiv2025

T2R-bench: A Benchmark for Generating Article-Level Reports from Real World Industrial Tables

Extensive research has been conducted to explore the capabilities of large language models (LLMs) in table reasoning. However, the essential task of transforming tables information into reports remains a significant challenge for industrial applications. This task is plagued by two critical issues: 1) the complexity and diversity of tables lead to suboptimal reasoning outcomes; and 2) existing table benchmarks lack the capacity to adequately assess the practical application of this task. To fill this gap, we propose the table-to-report task and construct a bilingual benchmark named T2R-bench, where the key information flow from the tables to the reports for this task. The benchmark comprises 457 industrial tables, all derived from real-world scenarios and encompassing 19 industry domains as well as 4 types of industrial tables. Furthermore, we propose an evaluation criteria to fairly measure the quality of report generation. The experiments on 25 widely-used LLMs reveal that even state-of-the-art models like Deepseek-R1 only achieves performance with 62.71 overall score, indicating that LLMs still have room for improvement on T2R-bench.

cs.CL↗

arXiv2025

Learning to Optimize Capacity Planning in Semiconductor Manufacturing

In manufacturing, capacity planning is the process of allocating production resources in accordance with variable demand. The current industry practice in semiconductor manufacturing typically applies heuristic rules to prioritize actions, such as future change lists that account for incoming machine and recipe dedications. However, while offering interpretability, heuristics cannot easily account for the complex interactions along the process flow that can gradually lead to the formation of bottlenecks. Here, we present a neural network-based model for capacity planning on the level of individual machines, trained using deep reinforcement learning. By representing the policy using a heterogeneous graph neural network, the model directly captures the diverse relationships among machines and processing steps, allowing for proactive decision-making. We describe several measures taken to achieve sufficient scalability to tackle the vast space of possible machine-level actions. Our evaluation results cover Intel's small-scale Minifab model and preliminary experiments using the popular SMT2020 testbed. In the largest tested scenario, our trained policy increases throughput and decreases cycle time by about 1.8% each.

cs.LG↗

arXiv2025

DiEP: Adaptive Mixture-of-Experts Compression through Differentiable Expert Pruning

Despite the significant breakthrough of Mixture-of-Experts (MoE), the increasing scale of these MoE models presents huge memory and storage challenges. Existing MoE pruning methods, which involve reducing parameter size with a uniform sparsity across all layers, often lead to suboptimal outcomes and performance degradation due to varying expert redundancy in different MoE layers. To address this, we propose a non-uniform pruning strategy, dubbed \textbf{Di}fferentiable \textbf{E}xpert \textbf{P}runing (\textbf{DiEP}), which adaptively adjusts pruning rates at the layer level while jointly learning inter-layer importance, effectively capturing the varying redundancy across different MoE layers. By transforming the global discrete search space into a continuous one, our method handles exponentially growing non-uniform expert combinations, enabling adaptive gradient-based pruning. Extensive experiments on five advanced MoE models demonstrate the efficacy of our method across various NLP tasks. Notably, \textbf{DiEP} retains around 92\% of original performance on Mixtral 8$\times$7B with only half the experts, outperforming other pruning methods by up to 7.1\% on the challenging MMLU dataset.

cs.CL↗