Recent advances in diffusion models have enabled the creation of deceptively real images, posing significant security risks when misused. In this study, we reveal that natural and synthetic images exhibit distinct differences in the high-frequency domains of their Fourier power spectra after undergoing iterative noise perturbations through an inverse multi-step denoising process, suggesting that such noise can provide additional discriminative information for identifying synthetic images. Based on this observation, we propose a novel detection method that amplifies these differences by progressively adding noise to the original images across multiple timesteps, and train an ensemble of classifiers on these noised images. To enhance human comprehension, we introduce an explanation generation and refinement module to identify flaws located in AI-generated images. Additionally, we construct two new datasets, GenHard and GenExplain, derived from the GenImage benchmark, providing detection samples of greater difficulty and high-quality rationales for fake images. Extensive experiments show that our method achieves state-of-the-art performance with 98.91% and 95.89% detection accuracy on regular and harder samples, increasing a minimal of 2.51% and 3.46% compared to baselines. Furthermore, our method also generalizes effectively to images generated by other diffusion models. Our code and datasets will be made publicly available.
Long-Context Transformer Models (LCTMs) are vital for real-world applications but suffer high computational costs due to attention's quadratic complexity. Block-sparse attention mitigates this by focusing computation on critical regions, yet existing methods struggle with balancing accuracy and efficiency due to costly block importance measurements. In this paper, we introduce XAttention, a plug-and-play framework that dramatically accelerates long-context inference in Transformers models using sparse attention. XAttention's key innovation is the insight that the sum of antidiagonal values (i.e., from the lower-left to upper-right) in the attention matrix provides a powerful proxy for block importance. This allows for precise identification and pruning of non-essential blocks, resulting in high sparsity and dramatically accelerated inference. Across comprehensive evaluations on demanding long-context benchmarks-including RULER and LongBench for language, VideoMME for video understanding, and VBench for video generation. XAttention achieves accuracy comparable to full attention while delivering substantial computational gains. We demonstrate up to 13.5x acceleration in attention computation. These results underscore XAttention's ability to unlock the practical potential of block sparse attention, paving the way for scalable and efficient deployment of LCTMs in real-world applications. Code is available at https://github.com/mit-han-lab/x-attention.
The rapid advancement of audio generation technologies has escalated the risks of malicious deepfake audio across speech, sound, singing voice, and music, threatening multimedia security and trust. While existing countermeasures (CMs) perform well in single-type audio deepfake detection (ADD), their performance declines in cross-type scenarios. This paper is dedicated to studying the alltype ADD task. We are the first to comprehensively establish an all-type ADD benchmark to evaluate current CMs, incorporating cross-type deepfake detection across speech, sound, singing voice, and music. Then, we introduce the prompt tuning self-supervised learning (PT-SSL) training paradigm, which optimizes SSL frontend by learning specialized prompt tokens for ADD, requiring 458x fewer trainable parameters than fine-tuning (FT). Considering the auditory perception of different audio types,we propose the wavelet prompt tuning (WPT)-SSL method to capture type-invariant auditory deepfake information from the frequency domain without requiring additional training parameters, thereby enhancing performance over FT in the all-type ADD task. To achieve an universally CM, we utilize all types of deepfake audio for co-training. Experimental results demonstrate that WPT-XLSR-AASIST achieved the best performance, with an average EER of 3.58% across all evaluation sets. The code is available online.
Moral stories are a time-tested vehicle for transmitting values, yet modern NLP lacks a large, structured corpus that couples coherent narratives with explicit ethical lessons. We close this gap with TF1-EN-3M, the first open dataset of three million English-language fables generated exclusively by instruction-tuned models no larger than 8B parameters. Each story follows a six-slot scaffold (character -> trait -> setting -> conflict -> resolution -> moral), produced through a combinatorial prompt engine that guarantees genre fidelity while covering a broad thematic space. A hybrid evaluation pipeline blends (i) a GPT-based critic that scores grammar, creativity, moral clarity, and template adherence with (ii) reference-free diversity and readability metrics. Among ten open-weight candidates, an 8B-parameter Llama-3 variant delivers the best quality-speed trade-off, producing high-scoring fables on a single consumer GPU (<24 GB VRAM) at approximately 13.5 cents per 1,000 fables. We release the dataset, generation code, evaluation scripts, and full metadata under a permissive license, enabling exact reproducibility and cost benchmarking. TF1-EN-3M opens avenues for research in instruction following, narrative intelligence, value alignment, and child-friendly educational AI, demonstrating that large-scale moral storytelling no longer requires proprietary giant models.
Large reasoning models (LRMs), such as OpenAI-o1 and DeepSeek-R1, demonstrate impressive long-horizon reasoning capabilities. However, their reliance on static internal knowledge limits their performance on complex, knowledge-intensive tasks and hinders their ability to produce comprehensive research reports requiring synthesis of diverse web information. To address this, we propose \textbf{WebThinker}, a deep research agent that empowers LRMs to autonomously search the web, navigate web pages, and draft research reports during the reasoning process. WebThinker integrates a \textbf{Deep Web Explorer} module, enabling LRMs to dynamically search, navigate, and extract information from the web when encountering knowledge gaps. It also employs an \textbf{Autonomous Think-Search-and-Draft strategy}, allowing the model to seamlessly interleave reasoning, information gathering, and report writing in real time. To further enhance research tool utilization, we introduce an \textbf{RL-based training strategy} via iterative online Direct Preference Optimization (DPO). Extensive experiments on complex reasoning benchmarks (GPQA, GAIA, WebWalkerQA, HLE) and scientific report generation tasks (Glaive) demonstrate that WebThinker significantly outperforms existing methods and strong proprietary systems. Our approach enhances LRM reliability and applicability in complex scenarios, paving the way for more capable and versatile deep research systems. The code is available at https://github.com/RUC-NLPIR/WebThinker.
Large language models (LLMs) have demonstrated remarkable capabilities across a wide range of tasks, yet they often refuse to answer legitimate queries-a phenomenon known as overrefusal. Overrefusal typically stems from over-conservative safety alignment, causing models to treat many reasonable prompts as potentially risky. To systematically understand this issue, we probe and leverage the models'safety decision boundaries to analyze and mitigate overrefusal. Our findings reveal that overrefusal is closely tied to misalignment at these boundary regions, where models struggle to distinguish subtle differences between benign and harmful content. Building on these insights, we present RASS, an automated framework for prompt generation and selection that strategically targets overrefusal prompts near the safety boundary. By harnessing steering vectors in the representation space, RASS efficiently identifies and curates boundary-aligned prompts, enabling more effective and targeted mitigation of overrefusal. This approach not only provides a more precise and interpretable view of model safety decisions but also seamlessly extends to multilingual scenarios.We have explored the safety decision boundaries of various LLMs and construct the MORBench evaluation set to facilitate robust assessment of model safety and helpfulness across multiple languages. Code and datasets will be released at https://anonymous.4open.science/r/RASS-80D3.
Model distillation has become essential for creating smaller, deployable language models that retain larger system capabilities. However, widespread deployment raises concerns about resilience to adversarial manipulation. This paper investigates vulnerability of distilled models to adversarial injection of biased content during training. We demonstrate that adversaries can inject subtle biases into teacher models through minimal data poisoning, which propagates to student models and becomes significantly amplified. We propose two propagation modes: Untargeted Propagation, where bias affects multiple tasks, and Targeted Propagation, focusing on specific tasks while maintaining normal behavior elsewhere. With only 25 poisoned samples (0.25% poisoning rate), student models generate biased responses 76.9% of the time in targeted scenarios - higher than 69.4% in teacher models. For untargeted propagation, adversarial bias appears 6x-29x more frequently in student models on unseen tasks. We validate findings across six bias types (targeted advertisements, phishing links, narrative manipulations, insecure coding practices), various distillation methods, and different modalities spanning text and code generation. Our evaluation reveals shortcomings in current defenses - perplexity filtering, bias detection systems, and LLM-based autorater frameworks - against these attacks. Results expose significant security vulnerabilities in distilled models, highlighting need for specialized safeguards. We propose practical design principles for building effective adversarial bias mitigation strategies.
Photorealistic style transfer (PST) enables real-world color grading by adapting reference image colors while preserving content structure. Existing methods mainly follow either approaches: generation-based methods that prioritize stylistic fidelity at the cost of content integrity and efficiency, or global color transformation methods such as LUT, which preserve structure but lack local adaptability. To bridge this gap, we propose Spatial Adaptive 4D Look-Up Table (SA-LUT), combining LUT efficiency with neural network adaptability. SA-LUT features: (1) a Style-guided 4D LUT Generator that extracts multi-scale features from the style image to predict a 4D LUT, and (2) a Context Generator using content-style cross-attention to produce a context map. This context map enables spatially-adaptive adjustments, allowing our 4D LUT to apply precise color transformations while preserving structural integrity. To establish a rigorous evaluation framework for photorealistic style transfer, we introduce PST50, the first benchmark specifically designed for PST assessment. Experiments demonstrate that SA-LUT substantially outperforms state-of-the-art methods, achieving a 66.7% reduction in LPIPS score compared to 3D LUT approaches, while maintaining real-time performance at 16 FPS for video stylization. Our code and benchmark are available at https://github.com/Ry3nG/SA-LUT
3D Gaussian Splatting (3DGS) has emerged as a powerful paradigm for real-time photorealistic novel view synthesis. However, its training-time densification remains largely heuristic, often resulting in redundant Gaussian proliferation and inaccurate geometric reconstruction. Existing methods predominantly rely on first-order screen-space gradients to trigger cloning and splitting, which fail to distinguish true geometric discontinuities from high-frequency textures or noisy residuals. To address this limitation, we propose ACD-GS, an Asymmetric Curvature-aware Densification framework for compact and geometry-aware 3D Gaussian optimization. Our key insight is that second-order depth responses provide a more reliable structural prior for density control, enabling asymmetric regulation of cloning and splitting to better capture geometric discontinuities. Building on this, we introduce a multi-view photometric consistency constraint to suppress transient single-view errors, together with a capacity-aware adaptive regulation strategy that balances densification and pruning during training. Unlike post-hoc compression or quantization-based approaches, ACD-GS directly prevents redundant Gaussian generation during training, yielding inherently compact representations. Extensive experiments demonstrate that our method reduces nearly 60% of Gaussian count and storage while maintaining competitive rendering fidelity, consistently outperforming existing methods across multiple datasets. Our code is available in https://waveviewer.github.io/ACD-GS/.
The biological definition of Alzheimer's disease (AD) relies on multi-modal neuroimaging, yet the clinical utility of positron emission tomography (PET) is limited by cost and radiation exposure, hindering early screening at preclinical or prodromal stages. While generative models offer a promising alternative by synthesizing PET from magnetic resonance imaging (MRI), achieving subject-specific precision remains a primary challenge. Here, we introduce DIReCT$++$, a Domain-Informed ReCTified flow model for synthesizing multi-tracer PET from MRI combined with fundamental clinical information. Our approach integrates a 3D rectified flow architecture to capture complex cross-modal and cross-tracer relationships with a domain-adapted vision-language model (BiomedCLIP) that provides text-guided, personalized generation using clinical scores and imaging knowledge. Extensive evaluations on multi-center datasets demonstrate that DIReCT$++$ not only produces synthetic PET images ($^{18}$F-AV-45 and $^{18}$F-FDG) of superior fidelity and generalizability but also accurately recapitulates disease-specific patterns. Crucially, combining these synthesized PET images with MRI enables precise personalized stratification of mild cognitive impairment (MCI), advancing a scalable, data-efficient tool for the early diagnosis and prognostic prediction of AD. The source code will be released on https://github.com/ladderlab-xjtu/DIReCT-PLUS.
Reinforcement-learning-trained agents for retrieval-augmented generation retrieve content (passages, knowledge-graph fragments, or hyperedges) and supervise training with outcome rewards on the final answer. Across these paradigms the reasoning route through retrieved content never leaves the language model's hidden state, conflating declarative knowledge with the procedural knowledge that recurs across queries of the same logical type; content-only methods consequently plateau as reasoning depth grows. We introduce PatternBloom, an agentic RAG framework that shifts the unit of retrieval from content (what to read) to logic (how to reason). A first RL stage scores the agent's constructed evidence graph by a frozen oracle's information gain through the Information-Density Reward (IDR), a size-normalized mutual-information surrogate. High-reward trajectories are distilled into a Graph Pattern Memory (GPM) of type-abstracted reasoning skeletons, and a second RL stage uses the Pattern-Augmented Reward (PAR) to couple policy to memory, turning procedural reasoning into a learnable, externalized structural prior. A 7B PatternBloom outperforms $27$ baselines by $+15.0$ Avg-OOD EM, with the gain compounding with reasoning depth precisely because procedural knowledge, once distilled, is reused rather than re-derived at every query. We release our code at https://github.com/EllenYiGe/PatternBloom.
CLIP과 같이 대조 학습 기반으로 훈련된 비전-언어 모델(VLM)은 이미지-텍스트 결합 표현 학습에서 괄목할 만한 발전을 이루었으나, 구성적 이해(compositional understanding)에서는 여전히 어려움을 겪고 있다. 이러한 모델들은 흔히 "단어 가방(bag-of-words)" 행동을 보이며, 객체 간 관계, 속성-객체 결합(attribute-object bindings), 단어 순서 의존성을 포착하는 데 취약하다. 이 한계는 최적화 시 전역적 단일 벡터 표현에 의존하는 것뿐만 아니라, 쌍을 이루는 이미지-텍스트 데이터에 내재된 풍부한 구성적 정보를 충분히 활용하고 모델링하지 못하는 데서 비롯된다. 본 연구에서는 MACCO(MAsked Compositional Concept MOdeling)를 제안한다. 이 프레임워크는 한 모달리티에서 구성적 개념을 마스킹하고, 다른 모달리티의 전체 문맥 정보를 조건으로 이를 재구성함으로써, 모델이 교차 모달 구성적 구조를 더 효과적으로 포착하고 정렬할 수 있게 한다. 이 과정을 지원하기 위해, 마스킹된 특징을 모달 간(inter-modally) 및 모달 내(intra-modally)에서 공동으로 정렬하고 정규화하는 두 가지 보조 목적 함수(auxiliary objectives)를 도입한다. 5개의 구성성 벤치마크에 대한 광범위한 실험 및 심층 분석을 통해, 본 접근법이 VLM의 구성성을 크게 향상시킬 뿐만 아니라 통사 구조와 언어 정보 포착 능력도 개선함을 입증한다. 또한 향상된 구성성은 텍스트-이미지 생성(text-to-image generation)과 멀티모달 대규모 언어 모델(multimodal LLM)에도 이점을 제공한다. 코드는 https://github.com/hiker-lw/MACCO에서 확인할 수 있다.
SproutRAG는 학습된 문장 간 어텐션으로 이진 청킹 트리를 구성해 문장 수준부터 다중 입도로 검색할 수 있는 계층적 RAG(검색-증강 생성) 프레임워크로, 추가 LLM 호출이나 손실 요약 없이 문맥 일관성과 검색 정밀도를 함께 확보한다. 과학·법률·오픈 도메인 4개 벤치마크에서 최강 baseline 대비 정보 효율성(IE)을 평균 6.1% 향상했다.
모호한 사용자 의도를 선호에 맞는 구조화·실행 가능 명세로 바꾸는 능력을 평가하는 벤치마크 SpecBench를 제안하고, 형태론적 분석(morphological analysis)으로 설계 차원을 분해한 뒤 시뮬레이션 사용자로 선택을 검토하고 실제 사용자와 협업하는 에이전트 Buddy를 소개한다. 기존 CLI 에이전트는 과신해 조기 구현에 들어가거나 질문 예산을 낭비하는 경향이 있었으나 Buddy는 더 적은 질의로 사용자 선호를 효과적으로 이끌어냈다.
MASTE는 리뷰 문장에서 (속성, 의견, 감성) 삼중항을 제로샷으로 추출하기 위해, 속성 추출·의견 추출·감성 추론·일관성 검증의 네 전문 에이전트를 순차적으로 조건화하는 멀티 에이전트 파이프라인을 제안한다. 학습 데이터 없이 네 가지 ASTE 벤치마크에서 기존 제로샷 및 사고의 연쇄 기반 LLM 베이스라인을 크게 앞서며 완전 지도학습 방법과의 격차를 좁혔다.
향후 10년의 관측 우주론은 Vera C. Rubin Observatory의 지상 기반 LSST와 우주 기반 Euclid 미션 같은 대규모 전천 탐사에 의해 형성될 것이다. 이 탐사들은 깊이, 해상도, 파장에 걸쳐 전례 없는 우주의 모습을 제공할 것으로 기대되지만, 관측 방식, 하늘 커버리지, 점퍼짐 함수(point-spread function), 스캔 주기의 차이로 인해 공동 분석이 유익하면서도 어렵다. 공동 분석을 용이하게 하기 위해, 우리는 지상 기반과 우주 기반 관측 사이를 변환하는 양방향 생성 모델인 A(stronomical)S(urvey)-Bridge를 소개한다. AS-Bridge는 LSST와 Euclid 관측 사이에 확률적 브라운 브리지(Brownian Bridge) 과정을 활용하는 확산 모델을 학습한다. 두 탐사는 하늘에서 겹치는 영역을 가지며, 이를 통해 두 탐사 간의 조건부 확률 분포를 명시적으로 모델링할 수 있다. 이 공식화가 단일 탐사 분석을 넘어 새로운 과학적 기능, 즉 누락된 탐사 관측에 대한 충실한 확률적 예측과 탐사 간 희귀 사건 검출을 가능하게 함을 보인다. 이러한 결과는 탐사 간 생성 모델링의 실현 가능성을 확립한다. 따라서 AS-Bridge는 미래의 LSST-Euclid 공동 데이터 파이프라인의 보완적 구성 요소로 기능하여, 두 탐사의 데이터가 이용 가능해지면 과학적 성과를 향상시킬 것으로 기대된다. 데이터와 코드는 https://github.com/ZHANG7DC/AS-Bridge에서 이용 가능하다.
The transition to sixth-generation (6G) networks transforms wireless infrastructure into a cognitive substrate supporting Vehicle-to-Everything (V2X), Industrial IoT (IIoT), and Integrated Sensing and Communication (ISAC). In this paradigm, autonomous agentic AI performs orchestration at millisecond scales, rendering traditional static governance frameworks fundamentally inadequate for risk management. This paper introduces GIRAF(Governance-Integrated Risk and Assurance Framework), a Governance-as-Code (GaC) framework for real-time risk quantification and trust modulation in agentic 6G systems. GIRAF derives a continuous Aggregate Risk Index ($R_{t}$) from machine-readable runtime signals, including epistemic confidence, network jitter, and verification latency. A core contribution is the formalization of the verification staleness trade-off, where safety mechanisms induce risk if computational latency exceeds 6G deadlines. We demonstrate that GIRAF identifies 'Confidence Gaps' discrepancies between agent reported certainty and environmental ground truth, triggering automated safety envelopes when conditions deteriorate. Crucially, GIRAF serves as the foundational governance groundwork and conceptual 'glue' that externalizes these technical risks into machine-readable telemetry. Through simulations with fine-tuned Large Language Models (LLMs), we validate that the framework preserves operational integrity while providing the essential actuarial baseline required for multi-stakeholder liability attribution and dynamic premium quantification in the 6G ecosystem.
대규모 언어 모델(LLM)의 최근 발전은 다단계 추론이 내부적으로 어떻게 표현되는지에 대한 관심을 새롭게 불러일으켰다. 기존 연구는 추론을 선형 체인으로 다루는 경우가 많지만, 많은 추론 문제는 중간 결론이 분기·병합·재사용되는 방향성 비순환 그래프(DAG)로 더 자연스럽게 모델링할 수 있다. 이러한 그래프 구조가 모델 내부에 반영되는지는 아직 불분명하다. 우리는 기저 추론 DAG의 속성이 LLM 은닉 상태에 선형적으로 인코딩되는지 검증하는 프레임워크인 Reasoning DAG Probing을 제안한다. 각 전제, 중간 결론, 최종 답을 DAG 노드로 형식화하고, 은닉 상태로부터 노드 깊이, 쌍별 거리, 인접 관계를 예측하는 경량 프로브를 학습한다. 이 프로브를 이용해 층별 DAG 구조의 출현을 분석하고, 근사 추론 그래프를 재구성하며, 추론 관련 구조를 교란하는 대조 조건을 평가한다. 논리·수학·코드 추론 벤치마크 전반에서 DAG 구조가 LLM 표현에 의미 있게 인코딩됨을 확인했다: 복원 가능성은 중간 층에서 정점을 이루고, 후반 층은 더 깊은 노드와 장거리 의존성을 우선적으로 복원하며, 더 큰 모델은 DAG 복원 가능성이 더 강하다. 자기회귀 생성 동안 사후학습(post-training)된 모델은 더 뚜렷하고 점진적으로 강화되는 DAG 구조를 보이며, 이는 전반적으로 답의 정확성 추세를 선행한다. 이 결과는 LLM 추론이 순수하게 순차적이지 않으며 측정 가능한 내부 그래프 구조를 드러냄을 시사한다.
Vision Transformers (ViTs) face two limitations: the rigid resolution dependency of positional embeddings, which complicates cross-resolution fine-tuning, and the quadratic complexity of attention. While Swin Transformer alleviates the latter through window attention, it suffers from fine-tuning. Following the philosophy "no token is an island," we present Iwin Transformer, a position-embedding-free hierarchical vision transformer that couples interleaved window attention with depthwise convolution inside a single block. Attention captures long-range dependencies, while convolution links local neighbors and implicitly encodes spatial position. This design not only reduces the quadratic complexity of attention but also enables two types of scalability: fine-tuning from low to high resolution and weight transfer from 2D to 3D. With window-size adjustment alone, direct $224^2{\rightarrow}384^2$ fine-tuning lifts Iwin-L from 86.4\% to 87.4\% top-1 accuracy on ImageNet-1K. Transferring an ImageNet-pretrained Iwin-T to video achieves 79.1\% on Kinetics-400, outperforming Swin-T (78.8\%) with 15.9\% fewer FLOPs. Iwin also remains competitive on ADE20K segmentation and class-conditional image generation (FlashDiT). Overall, Iwin offers an effective approach to simultaneously tackling the complexity and scalability challenges in ViTs. Code and models are at https://github.com/cominder/Iwin-Transformer.
State-of-the-art methods in image-to-image translation are capable of learning a mapping from a source domain to a target domain with unpaired image data. Though the existing methods have achieved promising results, they still produce visual artifacts, being able to translate low-level information but not high-level semantics of input images. One possible reason is that generators do not have the ability to perceive the most discriminative parts between the source and target domains, thus making the generated images low quality. In this paper, we propose a new Attention-Guided Generative Adversarial Networks (AttentionGAN) for the unpaired image-to-image translation task. AttentionGAN can identify the most discriminative foreground objects and minimize the change of the background. The attention-guided generators in AttentionGAN are able to produce attention masks, and then fuse the generation output with the attention masks to obtain high-quality target images. Accordingly, we also design a novel attention-guided discriminator which only considers attended regions. Extensive experiments are conducted on several generative tasks with eight public datasets, demonstrating that the proposed method is effective to generate sharper and more realistic images compared with existing competitive models. The code is available at https://github.com/Ha0Tang/AttentionGAN.