Federated Class-Incremental Learning (FCIL) enables Class-Incremental Learning (CIL) from distributed data. Existing FCIL methods typically integrate old knowledge preservation into local client training. However, these methods cannot avoid spatial-temporal client drift caused by data heterogeneity and often incur significant computational and communication overhead, limiting practical deployment. To address these challenges simultaneously, we propose a novel approach, Spatial-Temporal Statistics Aggregation (STSA), which provides a unified framework to aggregate feature statistics both spatially (across clients) and temporally (across stages). The aggregated feature statistics are unaffected by data heterogeneity and can be used to update the classifier in closed form at each stage. Additionally, we introduce STSA-E, a communication-efficient variant with theoretical guarantees, achieving similar performance to STSA-E with much lower communication overhead. Extensive experiments on three widely used FCIL datasets, with varying degrees of data heterogeneity, show that our method outperforms state-of-the-art FCIL methods in terms of performance, flexibility, and both communication and computation efficiency. The code is available at https://github.com/Yuqin-G/STSA.

With the rapid advancement of diffusion models, AI-generated content (AIGC), particularly in image and video generation, has achieved remarkable progress while simultaneously raising significant concerns about content safety within the community. Compared to image generation, the risks of generated videos remain largely overlooked and more challenging, as harmful content may be concealed at any temporal location or accumulate over time. Therefore, this paper proposes a novel Dual-Safety Approach (DSA) to remove temporal risks in generated videos while preserving their inherent quality at test time. The DSA solves the temporal risks from two aspects: the semantics in the text prompts and the temporal dynamics in the diffusion generation process. Specifically, we introduce a Semantic Adaptive Refinement module, which constructs a prompt-specific harmful subspace for targeted risk mitigation while preserving core semantics. Rather than treating video safety at a coarse, whole-video level, we introduce Temporal Dynamic Adaptive Optimization and Temporal Safety Guidance to dynamically constrain unsafe content over time while preserving temporal consistency. Moreover, to evaluate the effectiveness of our approach, we construct a novel Temporal Safety Benchmark for both Text-to-Video and Image-to-Video models. Extensive experiments demonstrate that our DSA can significantly reduce the risks of videos generated by open-source models without degrading video quality and even surpassing commercial models. Additional evaluations on multiple benchmarks, including adversarial attack datasets, further confirm that DSA surpasses existing methods in both effectiveness and robustness. Code and data will be publicly available.

Federated learning with LoRA is widely used for parameter-efficient fine-tuning of large language models. To address system heterogeneity, current approaches typically assign varying LoRA ranks to clients with different resources. However, these heterogeneous methods suffer from feature dilution and high communication overhead. Furthermore, they often overlook the inconsistent update directions caused by Non-IID data. To overcome these limitations, we propose a simple yet effective federated heterogeneous LoRA approach. Our method treats the LoRA’s A matrix as a fixed orthogonal encoder to pre-align low-rank subspaces. Clients exclusively learn the decoder matrix B, which enables efficient dimensionality aggregation. Experimental results demonstrate that our method significantly outperforms state-of-the-art approaches in both convergence speed and communication efficiency.

Content-aware layout generation is a critical technique for automating poster design, as it ensures the effective spatial organization of multimodal elements for enhanced clarity and impact. Existing methods face notable limitations: data-driven models depend on scarce clean-canvas datasets or noisy pseudo-canvases, while large language model–based approaches incur high cost and lack fine-grained visual sensitivity. We propose Stable Layout Image Diffusion (SLID), a diffusion-based framework that reformulates layout generation as image synthesis with color-coded representations, avoiding pseudo-canvas artifacts and enabling stable training. Content-aware conditioning with subject mask, edge map, and elements image further enhances spatial reasoning and supports diverse, irregular layouts. We also introduce DC-Poster, a dataset of 37k annotated posters. Experiments show SLID surpasses state-of-the-art methods, producing visually appealing and semantically aligned layouts suitable for real-application.

Diffusion models have been widely deployed in AIGC services, but their reliance on opaque training data exposes them to backdoor attacks. In practical auditing scenarios, auditors are typically unable to access model parameters due to intellectual property protection, making white-box or query-intensive detection impractical. After detection, existing detoxification approaches are trapped in a dilemma between detoxification effectiveness and generation quality for service providers. We reveal Temporal Noise Consistency (TNC), a previously unreported phenomenon in which backdoor activation disrupts the consistency of noise predictions between adjacent diffusion timesteps within specific temporal segments, while clean inputs remain stable. Based on this finding, we propose TNC-Defense, a closed-loop framework for gray-box backdoor detection and model repair. Specifically, TNC-Detect (for auditors) uses inference-stage adjacent-noise statistics to detect backdoors and precisely localize anomalous timesteps without model-weight access. TNC-Detox (for service providers) utilizes these locations to perform trigger-agnostic, timestep-aware correction of the generation path, suppressing backdoor behavior while reducing detoxification cost. Across five representative backdoor attacks and state-of-the-art defenses, TNC-Defense improves the average detection accuracy by 11% with negligible additional overhead, and invalidates an average of 98.5% of triggered samples with only a mild degradation in generation quality. Our code is publicly available at: https://github.com/binzhwang/TNC-Defense.

This paper investigates how to leverage emerging non-volatile memory (NVM) to enhance the performance of Log-Structure Merge (LSM) tree based key-value (KV) stores. We propose KVFG-DB, which efficiently integrates fine-block granularity and single-NVM-level compaction, to deliver high write and read performance with minimal write amplification and reduced write stalls. KVFG-DB leverages a streamlined SSTable layout called CSSTable to manage data blocks (each containing ordered KV pairs) and their indexes (i.e., the minimal and maximal keys of every block), optimizing both storage costs and compaction performance. It organizes the LSM-tree data as CSSTables into a single level on NVM, with the left area serving as a buffer to receive flushed data with substantially greater capacity, and the right area storing compacted data in a global order among CSSTables. A fine-grained compaction is then performed under various conditions to select the most relevant data blocks with intersecting keys from CSSTables in both areas, facilitating byte-addressable, fast parallel execution across multiple threads. As a result, KVFG-DB adaptively compacts data and quickly moves them from the left area to the right area to enhance performance efficiency, significantly reducing write amplification and further mitigating write stalls. Our extensive experimental studies demonstrate that KVFG-DB achieves 1.2 & times; and 2 & times; lower write amplification, compared with state-of-the-art KV stores MioDB and SLM-DB. Accordingly, KVFG-DB shows a 1.1 & times; and 7.1 & times; improvement in random write performance compared to them, with tail latency reduced by 1.1 & times; and 7 & times;, respectively.

The rapid advancement of deep learning has turned models into highly valuable assets due to their reliance on massive data and costly training processes. However, these models are increasingly vulnerable to leakage and theft, highlighting the critical need for robust intellectual property protection. Model watermarking has emerged as an effective solution, with black-box watermarking gaining significant attention for its practicality and flexibility. Nonetheless, existing black-box methods often fail to better balance covertness (hiding the watermark to prevent detection and forgery) and robustness (ensuring the watermark resists removal)-two essential properties for real-world copyright verification. In this paper, we propose ComMark, a novel black-box model watermarking framework that leverages frequency-domain transformations to generate compressed, covert, and attack-resistant watermark samples by filtering out high-frequency information. To further enhance watermark robustness, our method incorporates simulated attack scenarios and a similarity loss during training. Comprehensive evaluations across diverse datasets and architectures demonstrate that ComMark achieves state-of-the-art performance in both covertness and robustness. Furthermore, we extend its applicability beyond image recognition to tasks including speech recognition, sentiment analysis, image generation, image captioning, and video recognition, underscoring its versatility and broad applicability.

As high-value intellectual assets, graph signals demand robust protection of their ownership. While digital watermarking offers a solution, existing methods rely on brittle, handcrafted techniques that yield a suboptimal trade-off between robustness and transparency while entirely neglecting security. To overcome these limitations, we propose GS-mark, the first deep learning framework that formulates graph signal watermarking as an end-to-end adversarial learning task. At its core, a key–conditioned, permutation-equivariant generator provides intrinsic security and robustness, while a multi-objective adversarial loss expertly navigates the inherent trade-off between the watermark embedding robustness and transparency. Extensive experiments on benchmark datasets show that GS-mark achieves exceptional robustness with high transparency, verifiable security against forgery.

The believable simulation of multi-user behavior is crucial for understanding complex social systems. Recently, large language models (LLMs)-based AI agents have made significant progress, enabling them to achieve human-like intelligence across various tasks. However, real human societies are often dynamic and complex, involving numerous individuals engaging in multimodal interactions. In this paper, taking e-commerce scenarios as an example, we present LMAgent, a very large-scale and multimodal agents society based on multimodal LLMs. In LMAgent, besides freely chatting with friends, the agents can autonomously browse, purchase, and review products, even perform live streaming e-commerce. To simulate this complex system, we introduce a self-consistency prompting mechanism to augment agents' multimodal capabilities, resulting in significantly improved decision-making performance over the existing multi-agent system. Moreover, we propose a fast memory mechanism combined with the small-world model to enhance system efficiency, which supports more than 10,000 agent simulations in a society. Experiments on agents' behavior show that these agents achieve comparable performance to humans in behavioral indicators. Furthermore, compared with the existing LLMs-based multi-agent system, more different and valuable phenomena are exhibited, such as herd behavior, which demonstrates the potential of LMAgent in credible large-scale social behavior simulations.

CPU-based trusted execution environments (TEEs) and differential privacy (DP) have gained wide applications for private inference. Due to high inference latency in TEEs, researchers use partition-based approaches that offload linear model components to GPUs. However, dense nonlinear layers of large language models (LLMs) result in significant communication overhead between TEEs and GPUs. DP-based approaches apply random noise to protect data privacy, but this compromises LLM performance and semantic understanding. To overcome the above drawbacks, this paper proposes CMIF, a Confidential and efficient Model Inference Framework. CMIF confidentially deploys the embedding layer in the client-side TEE and subsequent layers on GPU servers. Meanwhile, it optimizes the Report-Noisy-Max mechanism to protect sensitive inputs with a slight decrease in model performance. Extensive experiments on Llama-series models demonstrate that CMIF reduces additional inference overhead in TEEs while preserving user data privacy.

Rumor propagation modeling is critical for understanding and mitigating misinformation. Existing approaches combining rule-based regular agents with LLM-driven core agents provide a promising paradigm for large-scale rumor simulation. However, overlooking the dynamic nature of core agents and the importance of network topology on rumor spread significantly undermines the simulation performance. To address these issues, we present RumorSphere, a dynamic and hierarchical resonance framework for effective rumor simulation at the million-agent scale. Considering the dynamic role of core agents in rumor evolution, we propose a multi-agent dynamic interaction strategy based on the information cocoon theory, which adaptively identifies and activates critical core agents at conflict boundaries using LLMs, effectively supporting simulations with millions of agents. In addition, we design a hierarchical resonance network that integrates opinion leaders and localized community structures, enabling more realistic modeling of explosive rumor spread in real-world scenarios. Experiments on real-world datasets show that RumorSphere outperforms state-of-the-art methods, reducing simulation bias by an average of 26.5

One key challenge in federated learning is the data heterogeneity of local distribution across clients, which makes local models converge to different optimal values, reducing the performance of the global model. An efficient way to alleviate the effects of Non-IID data is to train a generator and transmit it or the generated data to the server or clients to extend the training dataset. However, transferring the generator or generated data brings huge additional communication overhead. To address this issue, we propose a novel method, Federated Learning by Decoupling and Alleviating Knowledge Bias (FedDAKB). Specifically, we introduce an attention module to mask the critical elements of local feature maps and construct a simple generator to restore the masked elements by minimizing the distance between local feature maps and global class-specific feature maps. Compared to generator-based methods, our generator is more communication-efficient as it only introduces two simple convolution layers to recover masked elements. Then the global information is injected into the critical elements of each local feature map to alleviate bias in local feature extractors. Subsequently, the classifier bias can be alleviated by aligning global classifier logits with the ensemble logits of local classifiers with the global class-specific feature maps. Extensive experiments demonstrate that our method achieves state-of-the-art performance on Cifar-10, Cifar-100, and Tiny ImageNet.

Recent advancements in diffusion-based image editing pose a significant threat to the authenticity of digital visual content. Traditional embedding-based watermarking methods often introduce perceptible perturbations to maintain robustness, inevitably compromising visual fidelity. Meanwhile, existing zero-watermarking approaches, typically relying on global image features, struggle to withstand sophisticated manipulations. In this work, we reveal a critical insight: while individual image patches undergo substantial alterations during AI-based editing, the relational distance between patch pairs remain largely invariant. Leveraging this property, we propose Relational Zero-Watermarking (Rel-Zero), a novel framework that requires no modification to the original image but derives a unique zero-watermark from these editing-invariant patch relations. By grounding the watermark in intrinsic structural consistency rather than absolute appearance, Rel-Zero provides a non-invasive yet resilient mechanism for content authentication. Extensive experiments demonstrate that Rel-Zero achieves substantially improved robustness across diverse editing models and manipulations compared to prior zero-watermarking approaches.

Federated fine-tuning of Large Language Models (LLMs) faces significant challenges due to system heterogeneity across participating clients. While existing heterogeneous Low-Rank Adaptation (LoRA) methods attempt to address resource disparities by allocating variable ranks, they typically rely on zero-padding and truncation strategies. These approaches often suffer from aggregation noise and computational inefficiency. To overcome these limitations, we propose ProRes-LoRA, a Progressive Orthogonal Residual Decomposition framework. Unlike traditional dimension alignment approaches, ProRes-LoRA reformulates heterogeneous aggregation as a sequence of homogeneous residual learning subproblems. The training process is decoupled into two stages: "universal skeleton knowledge" and "refined texture learning". This mechanism effectively eliminates the optimization instability induced by sparse gradients in zero-padding and ensures feature space non-redundancy via orthogonal regularization. Extensive experiments demonstrate that ProRes-LoRA matches or exceeds the performance of HetLoRA while significantly reducing communication and computational overhead, achieving a 65.5% reduction compared to HetLoRA and 93.7% against FLoRA.

Graph-structured data is foundational to numerous web applications, and watermarking is crucial for protecting their intellectual property and ensuring data provenance. Existing watermarking methods primarily operate on graph structures or entangled graph representations, which compromise the transparency and robustness of watermarks due to the information coupling in representing graphs and uncontrollable discretization in transforming continuous numerical representations into graph structures. This motivates us to propose DRGW, the first graph watermarking framework that addresses these issues through disentangled representation learning. Specifically, we design an adversarially trained encoder that learns an invariant structural representation against diverse perturbations and derives a statistically independent watermark carrier, ensuring both robustness and transparency of watermarks. Meanwhile, we devise a graph-aware invertible neural network to provide a lossless channel for watermark embedding and extraction, guaranteeing high detectability and transparency of watermarks. Additionally, we develop a structure-aware editor that resolves the issue of latent modifications into discrete graph edits, ensuring robustness against structural perturbations. Experiments on diverse benchmark datasets demonstrate the superior effectiveness of DRGW.

Reliable watermarking of panoramic imagery is fundamentally challenged by arbitrary 3D rotations. As panoramas are defined on the sphere, they naturally transform under the action of SO(3), rendering conventional planar representations and augmentation-based robustness strategies inadequate and devoid of theoretical guarantees. To address this, we formulate panoramas as spherical signals and leverage SO(3) representation theory to derive provably rotation-invariant descriptors. While spherical harmonic coefficients transform equivariantly under rotations, the natural invariant constructions are typically limited to zeroth-order statistics which eliminate directional information and severely constrain embedding capacity. In this work, we introduce a principled third-order invariant construction by coupling higher-order SO(3) irreducible representations via tensor products and projecting onto the trivial representation. This yields a spherical invariant bispectrum that preserves phase information while remaining strictly rotation-invariant. Leveraging this property, we embed watermarks into higher-order spherical harmonic coefficients and recover them from invariant bispectral scalars, enabling reliable extraction under arbitrary 3D rotations. We provide a theoretical proof of SO(3) invariance for it and demonstrate experimentally its near-perfect robustness to continuous rotations while maintaining high visual fidelity.

As image data outsourcing to mobile cloud grows, data privacy has become a major concern. Privacy-preserving image retrieval aims to search over encrypted data without requiring decryption. However, existing schemes face three critical challenges: struggle to balance accuracy, efficiency, and security; limited scalability for large-scale image retrieval in multi-user settings; and vulnerability to security threats arising from user permission changes or key compromises. In this paper, we propose ELSEIR, a novel framework for accurate, efficient, and privacy-preserving image retrieval. ELSEIR leverages a deep hashing model to extract image feature vectors and designs an irreversible random hash code generation module that combines secure permutation keys with two differential privacy methods for privacy protection. To enable accurate searches in multi-user settings, ELSEIR introduces a key conversion protocol that allows the cloud to unify ciphertexts encrypted under different user keys via corresponding switch keys. Furthermore, we extend the framework to ABE-ELSEIR, which supports immediate and efficient user revocation. We further provide formal security proofs demonstrating that the proposed frameworks are resilient against known-plaintext and key collusion attacks. Extensive experiments on real-world datasets show that our scheme achieves accuracy comparable to the unprotected baseline, while surpassing existing approaches in both retrieval accuracy and efficiency.

The fine-tuning technique in deep learning gives rise to an emerging lineage relationship among models. This lineage provides a promising perspective for addressing security concerns such as unauthorized model redistribution and false claim of model provenance, which are particularly pressing in open-weight model libraries where robust lineage verification mechanisms are often lacking. Existing approaches to model lineage detection primarily rely on static architectural similarities, which are insufficient to capture the dynamic evolution of knowledge that underlies true lineage relationships. Drawing inspiration from the genetic mechanism of human evolution, we tackle the problem of model lineage attestation by verifying the joint trajectory of knowledge evolution and parameter modification. To this end, we propose a novel model lineage attestation framework. In our framework, model editing is first leveraged to quantify parameter-level changes introduced by fine-tuning. Subsequently, we introduce a novel knowledge vectorization mechanism that refines the evolved knowledge within the edited models into compact representations by the assistance of probe samples. The probing strategies are adapted to different types of model families. These embeddings serve as the foundation for verifying the arithmetic consistency of knowledge relationships across models, thereby enabling robust attestation of model lineage. Extensive experimental evaluations demonstrate the effectiveness and resilience of our approach in a variety of adversarial scenarios in the real world. Our method consistently achieves reliable lineage verification across a broad spectrum of model types, including classifiers, diffusion models, and large language models.

Manual formal analysis of cryptographic schemes is labor-intensive and requires substantial expertise. While model-checking tools (e.g., Scyther and Tamarin) and computational-security tools (e.g., CryptoVerif and EasyCrypt) improve the automation of security proofs, they still rely on experts to abstract schemes and write tool-specific formal descriptions. Large language models (LLMs) are a promising alternative, but their effectiveness in this domain remains unexplored due to the absence of standardized evaluation methodologies. To fill this gap, we introduce CrypFormBench (C.F.B for short), a comprehensive benchmark jointly covering symbolic and computational security to evaluate five core LLM capabilities: interpretation, generation, completion, transformation, and correction. It comprises 700 instances spanning 677 schemes, 7 mainstream formal verifier languages, and 160 security properties. The evaluation of 9 state-of-the-art LLMs reveals that most of them perform well on interpretation and completion, given their code-awareness advantages, but struggle with generation, transformation, and correction. Overall, their performance remains limited, with Claude-3.5 achieving the highest score at 48.7 out of 100. We further provide practical guidance, e.g., few-shot prompting, Pass@K sampling, and lightweight fine-tuning, to mitigate the executability bottleneck and improve tool-usable outputs. Taken together, our benchmark and analyses offer a grounded view of current progress and concrete directions toward reliable LLM-assisted formal cryptographic analysis.

Generative vision-language models like Stable Diffusion demonstrate remarkable capabilities in creative media synthesis, but they also pose substantial risks of producing unsafe, offensive, or culturally inappropriate content when prompted adversarially. Current defenses struggle to align outputs with human values without sacrificing generation quality or incurring high costs. To address these challenges, we introduce VALOR (Value-Aligned LLM-Overseen Rewriter), a modular, zero-shot agentic framework for safer and more helpful text-to-image generation. VALOR integrates layered prompt analysis with human-aligned value reasoning: a multi-level NSFW detector filters lexical and semantic risks; a cultural value alignment module identifies violations of social norms, legality, and representational ethics; and an intention disambiguator detects subtle or indirect unsafe implications. When unsafe content is detected, prompts are selectively rewritten by a large language model under dynamic, role-specific instructions designed to preserve user intent while enforcing alignment. If the generated image still fails a safety check, VALOR optionally performs a stylistic regeneration to steer the output toward a safer visual domain without altering core semantics. Experiments across adversarial, ambiguous, and value-sensitive prompts show that VALOR significantly reduces unsafe outputs by up to 100.00% while preserving prompt usefulness and creativity. These results highlight VALOR as a scalable and effective approach for deploying safe, aligned, and helpful image generation systems in open-world settings.