Diffusion Large Language Models (dLLMs) have recently emerged as a competitive non-autoregressive paradigm due to their unique training and inference approach. However, there is currently a lack of safety study on this novel architecture. In this paper, we present the first analysis of dLLMs' safety performance and propose a novel safety alignment method tailored to their unique generation characteristics. Specifically, we identify a critical asymmetry between the defender and attacker in terms of security. For the defender, we reveal that the middle tokens of the response, rather than the initial ones, are more critical to the overall safety of dLLM outputs; this seems to suggest that aligning middle tokens can be more beneficial to the defender. The attacker, on the contrary, may have limited power to manipulate middle tokens, as we find dLLMs have a strong tendency towards a sequential generation order in practice, forcing the attack to meet this distribution and diverting it from influencing the critical middle tokens. Building on this asymmetry, we introduce Middle-tOken Safety Alignment (MOSA), a novel method that directly aligns the model's middle generation with safe refusals exploiting reinforcement learning. We implement MOSA and compare its security performance against eight attack methods on two benchmarks. We also test the utility of MOSA-aligned dLLM on coding, math, and general reasoning. The results strongly prove the superiority of MOSA.
论文检索
输入标题、作者或关键词，从 9,256 篇学术成果中精准定位
Predicting the future popularity of information in online social networks is a crucial yet challenging task, due to the complex spatiotemporal dynamics underlying information diffusion. Existing methods typically use structural or sequential patterns within the observation window as direct inputs for subsequent popularity prediction. However, most approaches lack the ability to explicitly model the overall trend of popularity up to the prediction time, which leads to limited predictive capability. To address these limitations, we propose VNOIP, a novel method based on variational neural Ordinary Differential Equations (ODEs) for information popularity prediction. Specifically, VNOIP introduces bidirectional jump ODEs with attention mechanisms to capture long-range dependencies and bidirectional context within cascade sequences. Furthermore, by jointly considering both cascade patterns and overall trend temporal patterns, VNOIP explicitly models the continuous-time dynamics of popularity trend trajectories with variational neural ODEs. Additionally, a knowledge distillation loss is employed to align the evolution of prior and posterior latent variables. Extensive experiments on real-world datasets demonstrate that VNOIP is highly competitive in both prediction accuracy and efficiency compared to state-of-the-art baselines.
We address the challenge of integrating high-level semantic reasoning with low-level trajectory planning in end-to-end autonomous driving, where most existing frameworks decouple perception, decision-making, and control, leading to limited interpretability and poor instruction compliance. To bridge this gap, we propose Driving with Advice, a novel closed-loop framework that treats a vision-language model (VLM) as a motion advisor to provide interpretable, language-mediated guidance for trajectory generation. Our approach introduces three key innovations: (1) Semantic-Intentional Pretraining (SIP), which injects driving rationale into a compact VLM via machine-generated question-answering pairs; (2) a discrete action space grounded in directional and speed primitives, enabling structured and interpretable policy learning; and (3) an advice-following diffusion policy refined via Group Relative Policy Optimization under a multi-objective reward that ensures safety, comfort, and alignment with semantic intent. We evaluate our method on the NAVSIM benchmark in a closed-loop setting, achieving a state-of-the-art Predictive Driver Model Score (PDMS) of 91.5, outperforming strong baselines in safety (NC: 99.2). The results demonstrate that leveraging language as a cognitive interface between perception and control enhances both generalization and behavioral transparency, advancing the paradigm of language-conditioned driving.
Molecular structure generation from mass spectrometry is fundamental for understanding cellular metabolism and discovering novel compounds. Although tandem mass spectrometry (MS/MS) enables the high-throughput acquisition of fragment fingerprints, these spectra often reflect higher-order interactions involving the concerted cleavage of multiple atoms and bonds-crucial for resolving complex isomers and non-local fragmentation mechanisms. However, most existing methods adopt atom-centric and pairwise interaction modeling, overlooking higher-order edge interactions and lacking the capacity to systematically capture essential many-body characteristics for structure generation. To overcome these limitations, we present MBGen, a Many-Body enhanced diffusion framework for de novo molecular structure Generation from mass spectra. By integrating a many-body attention mechanism and higher-order edge modeling, MBGen comprehensively leverages the rich structural information encoded in MS/MS spectra, enabling accurate de novo generation and isomer differentiation for novel molecules. Experimental results on the NPLIB1 and MassSpecGym benchmarks demonstrate that MBGen achieves superior performance, with improvements of up to 230% over state-of-the-art methods, highlighting the scientific value and practical utility of many-body modeling for mass spectrometry-based molecular generation. Further analysis and ablation studies show that our approach effectively captures higher-order interactions and exhibits enhanced sensitivity to complex isomeric and non-local fragmentation information.
Computational fluid dynamics (CFD) simulations traditionally require extensive computational resources, limiting their utility in many scientific and engineering applications at scale. We introduce Physically-Informed Flow Matching Graph Networks (PIFM-GN), a novel generative framework that directly samples fluid states under specified physical conditions without requiring expensive time-stepping simulations. The key innovation of our approach is the incorporation of incompressibility constraints directly into the flow matching transport process by parameterizing velocity fields through vector potentials, with graph-based curl operators ensuring divergence-free predictions without requiring global pressure-Poisson solves. Experiments on diverse fluid dynamics problems -- ranging from two-dimensional surface pressure distributions and complete flow fields, to complex three-dimensional airflow fields -- demonstrate that PIFM-GN generates high-fidelity samples with significantly fewer sampling steps than diffusion-based alternatives. Most notably, our model maintains competitive performance even with a single sampling step, a regime where diffusion models completely fail. Our generated samples accurately reproduce the statistical characteristics of target flows, successfully capturing multi-modal pressure distributions across various flow conditions, while achieving significant computational speedups compared to diffusion-based methods. PIFM-GN thus enables efficient generation of fluid states for downstream analysis and design tasks in scientific and engineering applications.
Dual-Branch Asymmetric Discrepancy Learning Based on Fake Image Pattern-Coexistence for AI-Generated Image Detection
PDF ↗With the rapid advancement of generative models, high-fidelity AI-generated images have become increasingly indistinguishable from real images, posing significant challenges to traditional detection methods that rely on explicit artifacts or uniform feature learning. We hypothesize that detection ambiguity originates from pattern coexistence: synthetic images simultaneously embed (a) authentic patterns inherited from real-image distributions and (b) synthetic patterns induced by generative architectures, whereas real images maintain consistent patterns. We validate this hypothesis through SHAP-based quantitative analysis, demonstrating that synthetic images inherently exhibit a dual distribution—simultaneously containing authentic patterns and synthetic traces—while real images show a unimodal distribution. Building on this insight, this paper proposes a Dual-Branch Asymmetric Discrepancy Learning (DADL) framework. The DADL leverages multi-scale feature extraction and Asymmetric Feature Discrepancy Loss to capture and amplify such pattern differences across multiple scales. Extensive experiments on three benchmarks (AIGCDetectBenchmark, GenImage, and Chameleon) show that DADL achieves state-of-the-art performance, with particular strengths in detecting high-fidelity synthetic images from diffusion models (e.g., Midjourney, SDv1.4, SDv1.5) and enhancing generalization across diverse generative paradigms. This study not only offers an effective approach for AIGI detection but also sheds light on the intrinsic properties of synthetic images, providing a new perspective for advancing AIGI forensics.
RNA inverse folding, designing sequences to form specific 3D structures, is critical for therapeutics, gene regulation, and synthetic biology. Current methods, focused on sequence recovery, struggle to address structural objectives like secondary structure consistency (SS), minimum free energy (MFE), and local distance difference test (LDDT), leading to suboptimal structural accuracy. To tackle this, we propose a reinforcement learning (RL) framework integrated with a latent diffusion model (LDM). Drawing inspiration from the success of diffusion models in RNA inverse folding, which adeptly model complex sequence-structure interactions, we develop an LDM incorporating pre-trained RNA-FM embeddings from a large-scale RNA model. These embeddings capture co-evolutionary patterns, markedly improving sequence recovery accuracy. However, existing approaches, including diffusion-based methods, cannot effectively handle non-differentiable structural objectives. By contrast, RL excels in this task by using policy-driven reward optimization to navigate complex, non-gradient-based objectives, offering a significant advantage over traditional methods. In summary, we propose the Step-wise Optimization of Latent Diffusion Model (SOLD), a novel RL framework that optimizes single-step noise without sampling the full diffusion trajectory, achieving efficient refinement of multiple structural objectives. Experimental results demonstrate SOLD surpasses its LDM baseline and state-of-the-art methods across all metrics, establishing a robust framework for RNA inverse folding with profound implications for biotechnological and therapeutic applications.
Social media has fundamentally transformed how people access information and form social connections, with content expression playing a critical role in driving information diffusion. While prior research has focused largely on network structures and tipping point identification, it provides limited tools for automatically generating content tailored for virality within a specific audience. To fill this gap, we propose the novel task of Diffusion-Oriented Content Generation (DOCG) and introduce an information enhancement algorithm for generating content optimized for diffusion. Our method includes an influence indicator that enables content-level diffusion assessment without requiring access to network topology, and an information editor that employs reinforcement learning to explore interpretable editing strategies. The editor leverages generative models to produce semantically faithful, audience-aware textual or visual content. Experiments on real-world social media datasets and user study demonstrate that our approach significantly improves diffusion effectiveness while preserving the core semantics of the original content.
Accident anticipation is essential for proactive and safe autonomous driving, where even a brief advance warning can enable critical evasive actions. However, two key challenges hinder real-world deployment: (1) noisy or degraded sensory inputs from weather, motion blur, or hardware limitations, and (2) the need to issue timely yet reliable predictions that balance early alerts with false-alarm suppression. We propose a unified framework that integrates diffusion-based denoising with a time-aware actor-critic model to address these challenges. The diffusion module reconstructs noise-resilient image and object features through iterative refinement, preserving critical motion and interaction cues under sensor degradation. In parallel, the actor-critic architecture leverages long-horizon temporal reasoning and time-weighted rewards to determine the optimal moment to raise an alert, aligning early detection with reliability. Experiments on three benchmark datasets (DAD, CCD, A3D) demonstrate state-of-the-art accuracy and significant gains in mean time-to-accident, while maintaining robust performance under Gaussian and impulse noise. Qualitative analyses further show that our model produces earlier, more stable, and human-aligned predictions in both routine and highly complex traffic scenarios, highlighting its potential for real-world, safety-critical deployment.
Metasurfaces are ultrathin, engineered materials composed of nanostructures that manipulate light in ways unattainable by natural materials. Recent advances have leveraged computational optimization, machine learning, and deep learning to automate their design. However, existing approaches exhibit two fundamental limitations: (1) they often restrict the model to generating only a subset of design parameters, and (2) they rely on heavily downsampled spectral targets, which compromises both the novelty and accuracy of the resulting structures. The core challenge lies in developing a generative model capable of exploring a large, unconstrained design space while precisely capturing the intricate physical relationships between material parameters and their high-resolution spectral responses. In this paper, we introduce MetaDiT, a novel framework for high-fidelity metasurface design that addresses these limitations. Our approach leverages a robust spectrum encoder pretrained with contrastive learning, providing strong conditional guidance to a Diffusion Transformer-based backbone. Experiments demonstrate that MetaDiT outperforms existing baselines in spectral accuracy, we further validate our method through extensive ablation studies.
Generative modeling has transformed many fields, such as language and visual modeling, while its application in financial markets remains under-explored. As the minimal unit within a financial market is an order, order-flow modeling represents a fundamental generative financial task. However, current approaches often yield unsatisfactory fidelity in generating order flow, and their generation lacks controllability, thereby limiting their practical applications. In this paper, we formulate the challenge of controllable financial market generation, and propose a Diffusion Guided Meta Agent (DigMA) model to address it. Specifically, we employ a conditional diffusion model to capture the dynamics of the market state represented by time-evolving distribution parameters of the mid-price return rate and the order arrival rate, and we define a meta agent with financial economic priors to generate orders from the corresponding distributions. Extensive experimental results show that DigMA achieves superior controllability and generation fidelity. Moreover, we validate its effectiveness as a generative environment for downstream high-frequency trading tasks and its computational efficiency.
TIDE: Temporal-Aware Sparse Autoencoders for Interpretable Diffusion Transformers in Image Generation
PDF ↗Diffusion Transformers (DiTs) are a powerful yet underexplored class of generative models compared to U-Net-based diffusion architectures. We propose TIDE—Temporal-aware sparse autoencoders for Interpretable Diffusion transformErs—a framework designed to extract sparse, interpretable activation features across timesteps in DiTs. TIDE effectively captures temporally-varying representations and reveals that DiTs naturally learn hierarchical semantics (e.g., 3D structure, object class, and fine-grained concepts) during large-scale pretraining. Experiments show that TIDE enhances interpretability and controllability while maintaining reasonable generation quality, enabling applications such as safe image editing and style transfer.
VoiceCloak: A Multi-Dimensional Defense Framework Against Unauthorized Diffusion-Based Voice Cloning
PDF ↗Diffusion Models (DMs) have achieved remarkable success in realistic voice cloning (VC), while they also increase the risk of malicious misuse. Existing proactive defenses designed for traditional VC models aim to disrupt the forgery process, but they have been proven incompatible with DMs due to the intricate generative mechanisms of diffusion. To bridge this gap, we introduce VoiceCloak, a multi-dimensional proactive defense framework with the goal of obfuscating speaker identity and degrading perceptual quality in potential unauthorized VC. To achieve these goals, we conduct a focused analysis to identify specific vulnerabilities within DMs, allowing VoiceCloak to disrupt the cloning process by introducing adversarial perturbations into the reference audio. Specifically, to obfuscate speaker identity, VoiceCloak first targets speaker identity by distorting representation learning embeddings to maximize identity variation, which is guided by auditory perception principles. Additionally, VoiceCloak disrupts crucial conditional guidance processes, particularly attention context, thereby preventing the alignment of vocal characteristics that are essential for achieving convincing cloning. Then, to address the second objective, VoiceCloak introduces score magnitude amplification to actively steer the reverse trajectory away from the generation of high-quality speech. Noise-guided semantic corruption is further employed to disrupt structural speech semantics captured by DMs, degrading output quality. Extensive experiments highlight VoiceCloak's outstanding defense success rate against unauthorized diffusion-based voice cloning. Additional audio samples of VoiceCloak are available in demo pages.
Information diffusion prediction is crucial for understanding social network dynamics, yet existing methods often neglect user participation uncertainty. This oversight typically stems from an implicit participation homogeneity assumption, which treats all observed interactions as equally reliable propagation signals, leading to fragile inferred topologies and uncertainty contamination. To address this, we propose SIEVE, a novel framework employing two synergistic strategies. First, robust node representations are learned via controllable uncertainty injection coupled with associated contrastive learning, mitigating topological fragility. Second, an uncertainty-aware directed graph aggregation mechanism is introduced, which dynamically constructs asymmetric aggregation topologies with adaptive weighting, thereby suppressing uncertainty contamination. Experiments on four public datasets demonstrate that SIEVE significantly outperforms state-of-the-art methods, offering valuable insights for designing robust information diffusion prediction models.
Understanding brain function represents a fundamental goal in neuroscience, with critical implications for therapeutic interventions and neural engineering applications. Computational modeling provides a quantitative framework for accelerating this understanding, but faces a fundamental trade-off between computational efficiency and high-fidelity modeling. To address this limitation, we introduce a novel Energy-based Autoregressive Generation (EAG) framework that employs an energy-based transformer learning temporal dynamics in latent space through strictly proper scoring rules, enabling efficient generation with realistic population and single-neuron spiking statistics. Evaluation on synthetic Lorenz datasets and two Neural Latents Benchmark datasets (MC_Maze and Area2_bump) demonstrates that EAG achieves state-of-the-art generation quality with substantial computational efficiency improvements, particularly over diffusion-based methods. Beyond optimal performance, conditional generation applications show two capabilities: generalizing to unseen behavioral contexts and improving motor brain-computer interface decoding accuracy using synthetic neural data. These results demonstrate the effectiveness of energy-based modeling for neural population dynamics with applications in neuroscience research and neural engineering.
As diffusion probabilistic models (DPMs) become central to Generative AI (GenAI), understanding their memorization behavior is essential for evaluating risks such as data leakage, copyright infringement, and trustworthiness. While prior research finds conditional DPMs highly susceptible to data extraction attacks using explicit prompts, unconditional models are often assumed to be safe. We challenge this view by introducing Surrogate condItional Data Extraction (SIDE), a general framework that constructs data-driven surrogate conditions to enable targeted extraction from any DPM. Through extensive experiments on CIFAR-10, CelebA, ImageNet, and LAION-5B, we show that SIDE can successfully extract training data from so-called safe unconditional models, outperforming baseline attacks even on conditional models. Complementing these findings, we present a unified theoretical framework based on informative labels, demonstrating that all forms of conditioning, explicit or surrogate, amplify memorization. Our work redefines the threat landscape for DPMs, establishing precise conditioning as a fundamental vulnerability and setting a new, stronger benchmark for model privacy evaluation.
Transferring Causal Driving Patterns for Generalizable Traffic Simulation with Diffusion-Based Distillation
PDF ↗Traffic simulation is essential for validating the safety and reliability of autonomous driving systems, yet data-driven simulation methods often struggle with distribution shifts, limiting their generalizability across diverse datasets (domains). To address this, we present Causal Driving Pattern Transfer (CDPT), a novel two-stage knowledge distillation framework built upon diffusion model to enhance cross-domain generalizability. In Phase I, we implement hybrid self-distillation within the source domain by integrating feature-, response-, and contrastive-level distillation, which enables the model to decompose complex driving behaviors into their core causal components, including scene-conditioned driven patterns, multi-agent interaction dynamics and casual saliency. In Phase II, we introduce a continual distillation strategy: few-shot samples from the target domain are used to initiate generation of diverse synthetic scenarios, allowing the student model to continually adapt to novel environments without retraining on large-scale data. Extensive experiments demonstrate that CDPT achieves strong generalization in both open-loop and closed-loop simulations, effectively generating realistic, interaction-aware behaviors that are critical for scalable and reliable autonomous driving testing.
The discovery of novel proteins relies on sensitive protein identification, for which de novo peptide sequencing (DNPS) from mass spectra is a crucial approach. While deep learning has advanced DNPS, existing models inadequately enforce the fundamental mass consistency constraint—that a predicted peptide's mass must match the experimental measured precursor mass. Previous DNPS methods often treat this critical information as a simple input feature or use it in post-processing, leading to numerous implausible predictions that do not adhere to this fundamental physical property. To address this limitation, we introduce DiffuNovo, a novel regressor-guided diffusion model for de novo peptide sequencing that provides explicit peptide-level mass control. Our approach integrates the mass constraint at two critical stages: during training, a novel peptide-level mass loss guides model optimization, while at inference, regressor-based guidance from gradient-based updates in the latent space steers the generation to compel the predicted peptide adheres to the mass constraint. Comprehensive evaluations on established benchmarks demonstrate that DiffuNovo surpasses state-of-the-art methods in DNPS accuracy. Additionally, as the first DNPS model to employ a diffusion model as its core backbone, DiffuNovo leverages the powerful controllability of diffusion architecture and achieves a significant reduction in mass error, thereby producing much more physically plausible peptides. These innovations represent a substantial advancement toward robust and broadly applicable DNPS. The source code is available in the supplementary material.
In a globalized world, cultural elements from diverse origins frequently appear together within a single visual scene. We refer to these as culture mixing scenarios, yet how Large Vision-Language Models (LVLMs) perceive them remains underexplored. We investigate culture mixing as a critical challenge for LVLMs and examine how current models behave when cultural items from multiple regions appear together. To systematically analyze these behaviors, we construct CultureMix, a food Visual Question Answering (VQA) benchmark with 23k diffusion-generated, human-verified culture mixing images across four subtasks: (1) food-only, (2) food+food, (3) food+background, and (4) food+food+background. Evaluating 10 LVLMs, we find consistent failures to preserve individual cultural identities in mixed settings. Models show strong background reliance, with accuracy dropping 14% when cultural backgrounds are added to food-only baselines, and they produce inconsistent predictions for identical foods across different contexts. To address these limitations, we explore three robustness strategies. We find supervised fine-tuning using a diverse culture mixing dataset substantially improve model consistency and reduce background sensitivity. We call for increased attention to culture mixing scenarios as a critical step toward developing LVLMs capable of operating reliably in culturally diverse real-world environments.
Recent advances in diffusion models (DMs) have achieved exceptional visual quality in image editing tasks. However, the global denoising dynamics of DMs inherently conflate local editing targets with the full-image context, leading to unintended modifications in non-target regions. In this paper, we shift our attention beyond DMs and turn to Masked Generative Transformers (MGTs) as an alternative approach to tackle this challenge. By predicting multiple masked tokens rather than holistic refinement, MGTs exhibit a localized decoding paradigm that endows them with the inherent capacity to explicitly preserve non-relevant regions during the editing process. Building upon this insight, we introduce the first MGT-based image editing framework, termed EditMGT. We first demonstrate that MGT's cross-attention maps provide informative localization signals for localizing edit-relevant regions and devise a multi-layer attention consolidation scheme that refines these maps to achieve fine-grained and precise localization. On top of these adaptive localization results, we introduce region-hold sampling, which restricts token flipping within low-attention areas to suppress spurious edits, thereby confining modifications to the intended target regions and preserving the integrity of surrounding non-target areas. To train EditMGT, we construct Crisp-2M, a high-resolution (>1024) dataset spanning seven diverse editing categories. Without introducing additional parameters, we adapt a pre-trained text-to-image MGT into an image editing model through attention injection. Extensive experiments across four standard benchmarks demonstrate that, with fewer than 1B parameters, our model achieves state-of-the-art image similarity performance while enabling faster editing. Moreover, it delivers comparable or superior editing quality, with improvements of 3.6% and 17.6% on style change and style transfer tasks, respectively.