Zhiyuan Ma

Zhiyuan Ma, Yuxiang Wei, Yabin Zhang et al. · stanford

By leveraging the text-to-image diffusion priors, score distillation can synthesize 3D contents without paired text-3D training data. Instead of spending hours of online optimization per text prompt, recent studies have been focused on learning a text-to-3D generative network for amortizing multiple text-3D relations, which can synthesize 3D contents in seconds. However, existing score distillation methods are hard to scale up to a large amount of text prompts due to the difficulties in aligning pretrained diffusion prior with the distribution of rendered images from various text prompts. Current state-of-the-arts such as Variational Score Distillation finetune the pretrained diffusion model to minimize the noise prediction error so as to align the distributions, which are however unstable to train and will impair the model's comprehension capability to numerous text prompts. Based on the observation that the diffusion models tend to have lower noise prediction errors at earlier timesteps, we propose Asynchronous Score Distillation (ASD), which minimizes the noise prediction error by shifting the diffusion timestep to earlier ones. ASD is stable to train and can scale up to 100k prompts. It reduces the noise prediction error without changing the weights of pre-trained diffusion model, thus keeping its strong comprehension capability to prompts. We conduct extensive experiments across different 2D diffusion models, including Stable Diffusion and MVDream, and text-to-3D generators, including Hyper-iNGP, 3DConv-Net and Triplane-Transformer. The results demonstrate ASD's effectiveness in stable 3D generator training, high-quality 3D content synthesis, and its superior prompt-consistency, especially under large prompt corpus.

Zhiyuan Ma, Xiangyu Zhu, Guojun Qi et al.

Controllability, generalizability and efficiency are the major objectives of constructing face avatars represented by neural implicit field. However, existing methods have not managed to accommodate the three requirements simultaneously. They either focus on static portraits, restricting the representation ability to a specific subject, or suffer from substantial computational cost, limiting their flexibility. In this paper, we propose One-shot Talking face Avatar (OTAvatar), which constructs face avatars by a generalized controllable tri-plane rendering solution so that each personalized avatar can be constructed from only one portrait as the reference. Specifically, OTAvatar first inverts a portrait image to a motion-free identity code. Second, the identity code and a motion code are utilized to modulate an efficient CNN to generate a tri-plane formulated volume, which encodes the subject in the desired motion. Finally, volume rendering is employed to generate an image in any view. The core of our solution is a novel decoupling-by-inverting strategy that disentangles identity and motion in the latent code via optimization-based inversion. Benefiting from the efficient tri-plane representation, we achieve controllable rendering of generalized face avatar at $35$ FPS on A100. Experiments show promising performance of cross-identity reenactment on subjects out of the training set and better 3D consistency.

Yipu Wang, Yuheng Ji, Yuyang Liu et al.

Cross-view correspondence is a fundamental capability for spatial understanding and embodied AI. However, it is still far from being realized in Vision-Language Models (VLMs), especially in achieving precise point-level correspondence, which is crucial for precise affordance interaction. So we propose the Cross-View Point Correspondence (CVPC) task and CrossPoint-Bench, a comprehensive benchmark with hierarchical design, inspired by the human cognitive process of "perceive", "reason", and "correspond". Our evaluation shows the state-of-the-art models (e.g., Gemini-2.5-Pro) still fall far behind humans, with a gap of over 54.65% in overall accuracy, exposing a challenge in transitioning from coarse-grained judgement to fine-grained coordinate prediction. To address this problem, we construct CrossPoint-378K, a dataset with 378K question-answering pairs across 900 scenes, focused on actionable affordance regions that better reflect real-world manipulation and interaction scenarios. Furthermore, we propose CroPond that trained on the CrossPoint-378K dataset. Our CroPond achieves state-of-the-art performance on CrossPoint-Bench, surpassing Gemini-2.5-Pro by 39.7% accuracy, which offers a foundation for advancing future work on cross-view correspondence. The benchmark, dataset, and model are publicly available at https://github.com/WangYipu2002/CrossPoint.

Yongqi Dong, Kejia Chen, Zhiyuan Ma

Condition-based maintenance is becoming increasingly important in hydraulic systems. However, anomaly detection for these systems remains challenging, especially since that anomalous data is scarce and labeling such data is tedious and even dangerous. Therefore, it is advisable to make use of unsupervised or semi-supervised methods, especially for semi-supervised learning which utilizes unsupervised learning as a feature extraction mechanism to aid the supervised part when only a small number of labels are available. This study systematically compares semi-supervised learning methods applied for anomaly detection in hydraulic condition monitoring systems. Firstly, thorough data analysis and feature learning were carried out to understand the open-sourced hydraulic condition monitoring dataset. Then, various methods were implemented and evaluated including traditional stand-alone semi-supervised learning models (e.g., one-class SVM, Robust Covariance), ensemble models (e.g., Isolation Forest), and deep neural network based models (e.g., autoencoder, Hierarchical Extreme Learning Machine (HELM)). Typically, this study customized and implemented an extreme learning machine based semi-supervised HELM model and verified its superiority over other semi-supervised methods. Extensive experiments show that the customized HELM model obtained state-of-the-art performance with the highest accuracy (99.5%), the lowest false positive rate (0.015), and the best F1-score (0.985) beating other semi-supervised methods.

Zhiyuan Ma, Zeyuan Li, Zihao Qiu et al.

In real-world applications of noninvasive electroencephalography (EEG), specialized decoders often show limited generalizability across diverse tasks under subject-independent settings. One central challenge is that task-relevant EEG signals often follow different temporal organization patterns across tasks, while many existing methods rely on task-tailored architectural designs that introduce task-specific temporal inductive biases. This mismatch makes it difficult to adapt temporal modeling across tasks without changing the model configuration. To address these challenges, we propose DSAINet, an efficient dual-scale attentive interaction network for general EEG decoding. Specifically, DSAINet constructs shared spatiotemporal token representations from raw EEG signals and models diverse temporal dynamics through parallel convolutional branches at fine and coarse scales. The resulting representations are then adaptively refined by intra-branch attention to emphasize salient scale-specific patterns and by inter-branch attention to integrate task-relevant features across scales, followed by adaptive token aggregation to yield a compact representation for prediction. Extensive experiments on five downstream EEG decoding tasks across ten public datasets show that DSAINet consistently outperforms 13 representative baselines under strict subject-independent evaluation. Notably, this performance is achieved using the same architecture hyperparameters across datasets. Moreover, DSAINet achieves a favorable accuracy-efficiency trade-off with only about 77K trainable parameters and provides interpretable neurophysiological insights. The code is publicly available at https://github.com/zy0929/DSAINet.

Tianfu Wang, Leilei Ding, Ziyang Tao et al.

High-fidelity diagram creation requires the complex orchestration of semantic topology, visual styling, and spatial layout, posing a significant challenge for automated systems. Existing methods also suffer from a representation gap: pixel-based models often lack precise control, while code-based synthesis limits intuitive flexibility. To bridge this gap, we introduce EvoDiagram, an agentic framework that generates object-level editable diagrams via an intermediate canvas schema. EvoDiagram employs a coordinated multi-agent system to decouple semantic intent from rendering logic, resolving conflicts across heterogeneous design layers. Additionally, we propose a design knowledge evolution mechanism that distills execution traces into a hierarchical memory of domain guidelines, enabling agents to retrieve context-aware expertise adaptively. We further release CanvasBench, a benchmark consisting of both data and metrics for canvas-based diagramming. Extensive experiments demonstrate that EvoDiagram exhibits excellent performance and balance against baselines in generating editable, structurally consistent, and aesthetically coherent diagrams. Our code is available at https://github.com/AuraX-AI/EvoDiagram.

Xianzhen Luo, Jingyuan Zhang, Shiqi Zhou et al.

Evaluating and improving the security capabilities of code agents requires high-quality, executable vulnerability tasks. However, existing works rely on costly, unscalable manual reproduction and suffer from outdated data distributions. To address these, we present CVE-Factory, the first multi-agent framework to achieve expert-level quality in automatically transforming sparse CVE metadata into fully executable agentic tasks. Cross-validation against human expert reproductions shows that CVE-Factory achieves 95\% solution correctness and 96\% environment fidelity, confirming its expert-level quality. It is also evaluated on the latest realistic vulnerabilities and achieves a 66.2\% verified success. This automation enables two downstream contributions. First, we construct LiveCVEBench, a continuously updated benchmark of 190 tasks spanning 14 languages and 153 repositories that captures emerging threats including AI-tooling vulnerabilities. Second, we synthesize over 1,000 executable training environments, the first large-scale scaling of agentic tasks in code security. Fine-tuned Qwen3-32B improves from 5.3\% to 35.8\% on LiveCVEBench, surpassing Claude 4.5 Sonnet, with gains generalizing to Terminal Bench (12.5\% to 31.3\%). We open-source CVE-Factory, LiveCVEBench, Abacus-cve (fine-tuned model), training dataset, and leaderboard. All resources are available at https://github.com/livecvebench/CVE-Factory .

Zhiyuan Ma, Guoli Jia, Biqing Qi et al.

Recently, stable diffusion (SD) models have typically flourished in the field of image synthesis and personalized editing, with a range of photorealistic and unprecedented images being successfully generated. As a result, widespread interest has been ignited to develop and use various SD-based tools for visual content creation. However, the exposure of AI-created content on public platforms could raise both legal and ethical risks. In this regard, the traditional methods of adding watermarks to the already generated images (i.e. post-processing) may face a dilemma (e.g., being erased or modified) in terms of copyright protection and content monitoring, since the powerful image inversion and text-to-image editing techniques have been widely explored in SD-based methods. In this work, we propose a Safe and high-traceable Stable Diffusion framework (namely Safe-SD) to adaptively implant the graphical watermarks (e.g., QR code) into the imperceptible structure-related pixels during the generative diffusion process for supporting text-driven invisible watermarking and detection. Different from the previous high-cost injection-then-detection training framework, we design a simple and unified architecture, which makes it possible to simultaneously train watermark injection and detection in a single network, greatly improving the efficiency and convenience of use. Moreover, to further support text-driven generative watermarking and deeply explore its robustness and high-traceability, we elaborately design lambda sampling and encryption algorithm to fine-tune a latent diffuser wrapped by a VAE for balancing high-fidelity image synthesis and high-traceable watermark detection. We present our quantitative and qualitative results on two representative datasets LSUN, COCO and FFHQ, demonstrating state-of-the-art performance of Safe-SD and showing it significantly outperforms the previous approaches.

Ruihuang Li, Zhengqiang Zhang, Chenhang He et al.

Recent vision-language pre-training models have exhibited remarkable generalization ability in zero-shot recognition tasks. Previous open-vocabulary 3D scene understanding methods mostly focus on training 3D models using either image or text supervision while neglecting the collective strength of all modalities. In this work, we propose a Dense Multimodal Alignment (DMA) framework to densely co-embed different modalities into a common space for maximizing their synergistic benefits. Instead of extracting coarse view- or region-level text prompts, we leverage large vision-language models to extract complete category information and scalable scene descriptions to build the text modality, and take image modality as the bridge to build dense point-pixel-text associations. Besides, in order to enhance the generalization ability of the 2D model for downstream 3D tasks without compromising the open-vocabulary capability, we employ a dual-path integration approach to combine frozen CLIP visual features and learnable mask features. Extensive experiments show that our DMA method produces highly competitive open-vocabulary segmentation performance on various indoor and outdoor tasks.

Tianyu Wang, Zhiyuan Ma, Qian Wang et al.

Recent advancements in image generation have achieved impressive results in producing high-quality images. However, existing image generation models still generally struggle with a spatial reasoning dilemma, lacking the ability to accurately capture fine-grained spatial relationships from the prompt and correctly generate scenes with structural integrity. To mitigate this dilemma, we propose RL-RIG, a Reinforcement Learning framework for Reflection-based Image Generation. Our architecture comprises four primary components: Diffuser, Checker, Actor, and Inverse Diffuser, following a Generate-Reflect-Edit paradigm to spark the Chain of Thought reasoning ability in image generation for addressing the dilemma. To equip the model with better intuition over generation trajectories, we further develop Reflection-GRPO to train the VLM Actor for edit prompts and the Image Editor for better image quality under a given prompt, respectively. Unlike traditional approaches that solely produce visually stunning yet structurally unreasonable content, our evaluation metrics prioritize spatial accuracy, utilizing Scene Graph IoU and employing a VLM-as-a-Judge strategy to assess the spatial consistency of generated images on LAION-SG dataset. Experimental results show that RL-RIG outperforms existing state-of-the-art open-source models by up to 11% in terms of controllable and precise spatial reasoning in image generation.

Yuxin Zuo, Kaiyan Zhang, Li Sheng et al. · pku, tsinghua

This paper investigates Reinforcement Learning (RL) on data without explicit labels for reasoning tasks in Large Language Models (LLMs). The core challenge of the problem is reward estimation during inference while not having access to ground-truth information. While this setting appears elusive, we find that common practices in Test-Time Scaling (TTS), such as majority voting, yield surprisingly effective rewards suitable for driving RL training. In this work, we introduce Test-Time Reinforcement Learning (TTRL), a novel method for training LLMs using RL on unlabeled data. TTRL enables self-evolution of LLMs by utilizing the priors in the pre-trained models. Our experiments demonstrate that TTRL consistently improves performance across a variety of tasks and models. Notably, TTRL boosts the pass@1 performance of Qwen-2.5-Math-7B by approximately 211% on the AIME 2024 with only unlabeled test data. Furthermore, although TTRL is only supervised by the maj@n metric, TTRL has demonstrated performance to consistently surpass the upper limit of the initial model maj@n, and approach the performance of models trained directly on test data with ground-truth labels. Our experimental findings validate the general effectiveness of TTRL across various tasks and highlight TTRL's potential for broader tasks and domains. GitHub: https://github.com/PRIME-RL/TTRL

Yongqi Dong, Kejia Chen, Yinxuan Peng et al.

As the central nerve of the intelligent vehicle control system, the in-vehicle network bus is crucial to the security of vehicle driving. One of the best standards for the in-vehicle network is the Controller Area Network (CAN bus) protocol. However, the CAN bus is designed to be vulnerable to various attacks due to its lack of security mechanisms. To enhance the security of in-vehicle networks and promote the research in this area, based upon a large scale of CAN network traffic data with the extracted valuable features, this study comprehensively compared fully-supervised machine learning with semi-supervised machine learning methods for CAN message anomaly detection. Both traditional machine learning models (including single classifier and ensemble models) and neural network based deep learning models are evaluated. Furthermore, this study proposed a deep autoencoder based semi-supervised learning method applied for CAN message anomaly detection and verified its superiority over other semi-supervised methods. Extensive experiments show that the fully-supervised methods generally outperform semi-supervised ones as they are using more information as inputs. Typically the developed XGBoost based model obtained state-of-the-art performance with the best accuracy (98.65%), precision (0.9853), and ROC AUC (0.9585) beating other methods reported in the literature.

Yabin Zhang, Wenjie Zhu, Hui Tang et al. · stanford

With the emergence of pre-trained vision-language models like CLIP, how to adapt them to various downstream classification tasks has garnered significant attention in recent research. The adaptation strategies can be typically categorized into three paradigms: zero-shot adaptation, few-shot adaptation, and the recently-proposed training-free few-shot adaptation. Most existing approaches are tailored for a specific setting and can only cater to one or two of these paradigms. In this paper, we introduce a versatile adaptation approach that can effectively work under all three settings. Specifically, we propose the dual memory networks that comprise dynamic and static memory components. The static memory caches training data knowledge, enabling training-free few-shot adaptation, while the dynamic memory preserves historical test features online during the testing process, allowing for the exploration of additional data insights beyond the training set. This novel capability enhances model performance in the few-shot setting and enables model usability in the absence of training data. The two memory networks employ the same flexible memory interactive strategy, which can operate in a training-free mode and can be further enhanced by incorporating learnable projection layers. Our approach is tested across 11 datasets under the three task settings. Remarkably, in the zero-shot scenario, it outperforms existing methods by over 3\% and even shows superior results against methods utilizing external training data. Additionally, our method exhibits robust performance against natural distribution shifts. Codes are available at \url{https://github.com/YBZh/DMN}.

Kaiyan Zhang, Yuxin Zuo, Bingxiang He et al. · pku, tsinghua

In this paper, we survey recent advances in Reinforcement Learning (RL) for reasoning with Large Language Models (LLMs). RL has achieved remarkable success in advancing the frontier of LLM capabilities, particularly in addressing complex logical tasks such as mathematics and coding. As a result, RL has emerged as a foundational methodology for transforming LLMs into LRMs. With the rapid progress of the field, further scaling of RL for LRMs now faces foundational challenges not only in computational resources but also in algorithm design, training data, and infrastructure. To this end, it is timely to revisit the development of this domain, reassess its trajectory, and explore strategies to enhance the scalability of RL toward Artificial SuperIntelligence (ASI). In particular, we examine research applying RL to LLMs and LRMs for reasoning abilities, especially since the release of DeepSeek-R1, including foundational components, core problems, training resources, and downstream applications, to identify future opportunities and directions for this rapidly evolving area. We hope this review will promote future research on RL for broader reasoning models. Github: https://github.com/TsinghuaC3I/Awesome-RL-for-LRMs

Yuxuan Sun, Yuze Zhao, Yufeng Wang et al.

Evaluating software engineering capabilities has become a core component of modern large language models (LLMs); however, the key bottleneck hindering further scaling lies not in the scarcity of high-quality solutions, but in the lack of high-quality test suites. Test suites are indispensable both for synthesizing program repair trajectories and for providing precise feedback signals in reinforcement learning. Unfortunately, due to the high cost and difficulty of annotation, high-quality test suites have long been hard to obtain, while those automatically generated by LLMs tend to be superficial and lack sufficient discriminative power. As a first step toward constructing high-quality test suites, we introduce SWE-Mutation, a benchmark for evaluating LLM-generated test suites. The benchmark characterizes test suites by introducing systematically mutated solutions that attempt to ``fool'' the test suites and pass validation. We further propose an agentic, language-agnostic framework for automatically generating complex mutants. Our benchmark consists of 2,636 mutated variants derived from 800 original instances and includes a multilingual subset spanning nine programming languages. Experiments on seven LLMs reveal that even DeepSeek-V3.1 achieves only 10.20% verification and 36.15% detection rates, highlighting the inadequacy of current LLMs. Additionally, our agentic mutation strategy enhances realism, reducing average detection rates from 71.04% to 39.81% compared to conventional methods. These findings expose persistent deficiencies in the ability of current LLMs to generate reliable and discriminative test suites.

Xiaofan Li, Ming Yang, Zhiyuan Ma et al.

Reinforcement Learning with Verifiable Rewards (RLVR) has catalyzed significant advances in the reasoning capabilities of Large Language Models (LLMs). However, effectively managing the exploration and exploitation trade-off remains a critical challenge. In this paper, we fully analyze the exploration and exploitation dilemma of extremely hard and easy samples during the training and propose a new fine-grained trade-off mechanism. Concretely, we introduce a perplexity space disentangling strategy that divides the sample space into distinct exploration (high perplexity) and exploitation (low perplexity) subspaces, thereby mining fine-grained samples requiring exploration-exploitation trade-off. Subsequently, we propose a bidirectional reward allocation mechanism with a minimum impact on verification rewards to implement perplexity-guided exploration and exploitation, enabling more stable policy optimization. Finally, we have evaluated our method on two mainstream tasks: mathematical reasoning and function calling, and experimental results demonstrate the superiority of the proposed method, confirming its effectiveness in enhancing LLM performance by fine-grained exploration-exploitation trade-off.

Lingchen Sun, Rongyuan Wu, Zhiyuan Ma et al.

Diffusion prior-based methods have shown impressive results in real-world image super-resolution (SR). However, most existing methods entangle pixel-level and semantic-level SR objectives in the training process, struggling to balance pixel-wise fidelity and perceptual quality. Meanwhile, users have varying preferences on SR results, thus it is demanded to develop an adjustable SR model that can be tailored to different fidelity-perception preferences during inference without re-training. We present Pixel-level and Semantic-level Adjustable SR (PiSA-SR), which learns two LoRA modules upon the pre-trained stable-diffusion (SD) model to achieve improved and adjustable SR results. We first formulate the SD-based SR problem as learning the residual between the low-quality input and the high-quality output, then show that the learning objective can be decoupled into two distinct LoRA weight spaces: one is characterized by the $\ell_2$-loss for pixel-level regression, and another is characterized by the LPIPS and classifier score distillation losses to extract semantic information from pre-trained classification and SD models. In its default setting, PiSA-SR can be performed in a single diffusion step, achieving leading real-world SR results in both quality and efficiency. By introducing two adjustable guidance scales on the two LoRA modules to control the strengths of pixel-wise fidelity and semantic-level details during inference, PiSASR can offer flexible SR results according to user preference without re-training. Codes and models can be found at https://github.com/csslc/PiSA-SR.

Zhiyuan Ma, Yuzhu Zhang, Guoli Jia et al.

As one of the most popular and sought-after generative models in the recent years, diffusion models have sparked the interests of many researchers and steadily shown excellent advantage in various generative tasks such as image synthesis, video generation, molecule design, 3D scene rendering and multimodal generation, relying on their dense theoretical principles and reliable application practices. The remarkable success of these recent efforts on diffusion models comes largely from progressive design principles and efficient architecture, training, inference, and deployment methodologies. However, there has not been a comprehensive and in-depth review to summarize these principles and practices to help the rapid understanding and application of diffusion models. In this survey, we provide a new efficiency-oriented perspective on these existing efforts, which mainly focuses on the profound principles and efficient practices in architecture designs, model training, fast inference and reliable deployment, to guide further theoretical research, algorithm migration and model application for new scenarios in a reader-friendly way. \url{https://github.com/ponyzym/Efficient-DMs-Survey}

Yanjun Guo, Zhengqiang Zhang, Pengfei Wang et al.

Spatially consistent long-horizon video generation aims to maintain temporal and spatial consistency along predefined camera trajectories. Existing methods mostly entangle memory modeling with video generation, leading to inconsistent content during scene revisits and diminished generative capacity when exploring novel regions, even trained on extensive annotated data. To address these limitations, we propose a decoupled framework that separates memory conditioning from generation. Our approach significantly reduces training costs while simultaneously enhancing spatial consistency and preserving the generative capacity for novel scene exploration. Specifically, we employ a lightweight, independent memory branch to learn precise spatial consistency from historical observation. We first introduce a hybrid memory representation to capture complementary temporal and spatial cues from generated frames, then leverage a per-frame cross-attention mechanism to ensure each frame is conditioned exclusively on the most spatially relevant historical information, which is injected into the generative model to ensure spatial consistency. When generating new scenes, a camera-aware gating mechanism is proposed to mediate the interaction between memory and generation modules, enabling memory conditioning only when meaningful historical references exist. Compared with the existing method, our method is highly data-efficient, yet the experiments demonstrate that our approach achieves state-of-the-art performance in terms of both visual quality and spatial consistency.

Yu Liu, Wenxiao Zhang, Cong Cao et al.

Answering real-world open-domain multi-hop questions over massive corpora is a critical challenge in Retrieval-Augmented Generation (RAG) systems. Recent research employs reinforcement learning (RL) to end-to-end optimize the retrieval-augmented reasoning process, directly enhancing its capacity to resolve complex queries. However, reliable deployment is hindered by two obstacles. 1) Retrieval Collapse: iterative retrieval over large corpora fails to locate intermediate evidence containing bridge answers without reasoning-guided planning, causing downstream reasoning to collapse. 2) Learning Instability: end-to-end trajectory training suffers from weak credit assignment across reasoning chains and poor error localization across modules, causing overfitting to benchmark-specific heuristics that limit transferability and stability. To address these problems, we propose PRISMA, a decoupled RL-guided framework featuring a Plan-Retrieve-Inspect-Solve-Memoize architecture. PRISMA's strength lies in reasoning-guided collaboration: the Inspector provides reasoning-based feedback to refine the Planner's decomposition and fine-grained retrieval, while enforcing evidence-grounded reasoning in the Solver. We optimize individual agent capabilities via Two-Stage Group Relative Policy Optimization (GRPO). Stage I calibrates the Planner and Solver as specialized experts in planning and reasoning, while Stage II utilizes Observation-Aware Residual Policy Optimization (OARPO) to enhance the Inspector's ability to verify context and trigger targeted recovery. Experiments show that PRISMA achieves state-of-the-art performance on ten benchmarks and can be deployed efficiently in real-world scenarios.

Zhiyuan Ma, Jiayu Liu, Xianzhen Luo et al.

Empowering large language models (LLMs) with effective tool utilization capabilities is crucial for enabling AI agents to solve complex problems. However, current models face two major limitations: (1) unreliable tool planning and invocation due to low-quality instruction datasets (e.g., widespread hallucinated API calls), and (2) weak tool reflection abilities (over 90% of errors cannot be corrected) resulting from static imitation learning. To address these critical limitations, we propose Tool-MVR, a novel Tool-Augmented LLM that achieves comprehensive System 2 reasoning through two key innovations. Specifically, we first introduce Multi-Agent Meta-Verification (MAMV), a systematic pipeline that rigorously validates APIs, queries, and reasoning trajectories to construct ToolBench-V, a new high-quality instruction dataset that addresses the limitation of unreliable tool planning and invocation. Second, we propose Exploration-based Reflection Learning (EXPLORE), which enhances tool reflection capabilities by leveraging tool feedback through a dynamic "Error -> Reflection -> Correction" learning paradigm, resulting in our reflection dataset ToolBench-R and addressing the critical weakness in tool reflection. Finally, we obtain Tool-MVR by finetuning open-source LLMs (e.g., Qwen-7B) on both ToolBench-V and ToolBench-R. Our experiments demonstrate that Tool-MVR achieves state-of-the-art performance on StableToolBench, surpassing both ToolLLM (by 23.9%) and GPT-4 (by 15.3%) while reducing API calls by 31.4%, with strong generalization capabilities across unseen tools and scenarios. Additionally, on our proposed RefineToolBench, the first benchmark specifically designed to evaluate tool reflection capabilities, Tool-MVR achieves a 58.9% error correction rate, significantly outperforming ToolLLM's 9.1%.

Pengfei Wang, Liyi Chen, Zhiyuan Ma et al.

Generating explorable 3D scenes from a single image is a highly challenging problem in 3D vision. Existing methods struggle to support free exploration, often producing severe geometric distortions and noisy artifacts when the viewpoint moves far from the original perspective. We introduce \textbf{One2Scene}, an effective framework that decomposes this ill-posed problem into three tractable sub-tasks to enable immersive explorable scene generation. We first use a panorama generator to produce anchor views from a single input image as initialization. Then, we lift these 2D anchors into an explicit 3D geometric scaffold via a generalizable, feed-forward Gaussian Splatting network. Instead of treating the panorama as a single image for reconstruction, we project it into multiple sparse anchor views and reformulate the reconstruction task as multi-view stereo matching, which allows us to leverage robust geometric priors learned from large-scale multi-view datasets. A bidirectional feature fusion module is used to enforce cross-view consistency, yielding an efficient and geometrically reliable scaffold. Finally, the scaffold serves as a strong prior for a novel view generator to produce photorealistic and geometrically accurate views at arbitrary cameras. By explicitly conditioning on a 3D-consistent scaffold to perform reconstruction, One2Scene works stably under large camera motions, supporting immersive scene exploration. Extensive experiments show that One2Scene substantially outperforms state-of-the-art methods in panorama depth estimation, feed-forward 360° reconstruction, and explorable 3D scene generation. Code and models will be released.

Shichao Ma, Zhiyuan Ma, Ming Yang et al.

Multi-turn tool-integrated reasoning enables Large Language Models (LLMs) to solve complex tasks through iterative information retrieval. However, current reinforcement learning (RL) frameworks for search-augmented reasoning predominantly rely on sparse outcome-level rewards, leading to a "Double Homogenization Dilemma." This manifests as (1) Process homogenization, where the thinking, reasoning, and tooling involved in generation are ignored. (2) Intra-group homogenization, coarse-grained outcome rewards often lead to inefficiencies in intra-group advantage estimation with methods like Group Relative Policy Optimization (GRPO) during sampling. To address this, we propose Turn-level Stage-aware Policy Optimization (TSPO). TSPO introduces the First-Occurrence Latent Reward (FOLR) mechanism, allocating partial rewards to the step where the ground-truth answer first appears, thereby preserving process-level signals and increasing reward variance within groups without requiring external reward models or any annotations. Extensive experiments demonstrate that TSPO significantly outperforms state-of-the-art baselines, achieving average performance gains of 24% and 13.6% on Qwen2.5-3B and 7B models, respectively.

Zhiyuan Ma, Ruixun Liu, Sixian Liu et al.

Recently, the rectified flow (RF) has emerged as the new state-of-the-art among flow-based diffusion models due to its high efficiency advantage in straight path sampling, especially with the amazing images generated by a series of RF models such as Flux 1.0 and SD 3.0. Although a straight-line connection between the noisy and natural data distributions is intuitive, fast, and easy to optimize, it still inevitably leads to: 1) Diversity concerns, which arise since straight-line paths only cover a fairly restricted sampling space. 2) Multi-scale noise modeling concerns, since the straight line flow only needs to optimize the constant velocity field $\bm v$ between the two distributions $\bmπ_0$ and $\bmπ_1$. In this work, we present Discretized-RF, a new family of rectified flow (also called momentum flow models since they refer to the previous velocity component and the random velocity component in each diffusion step), which discretizes the straight path into a series of variable velocity field sub-paths (namely ``momentum fields'') to expand the search space, especially when close to the distribution $p_\text{noise}$. Different from the previous case where noise is directly superimposed on $\bm x$, we introduce noise on the velocity $\bm v$ of the sub-path to change its direction in order to improve the diversity and multi-scale noise modeling abilities. Experimental results on several representative datasets demonstrate that learning momentum flow matching by sampling random velocity fields will produce trajectories that are both diverse and efficient, and can consistently generate high-quality and diverse results. Code is available at https://github.com/liuruixun/momentum-fm.

Chengrui Huang, Junshuo Zhang, Zhiyuan Ma et al.

Enabling large language models to scale and reliably use hundreds of tools is critical for real-world applications, yet challenging due to the inefficiency and error accumulation inherent in flat tool-calling architectures. To address this, we propose Hybrid Toolset Agentization & Adaptation (HTAA), a hierarchical framework for scalable tool-use planning. We propose a novel toolset agentization paradigm, which encapsulates frequently co-used tools into specialized agent tools, thereby reducing the planner's action space and mitigating redundancy. To ensure effective coordination, we design Asymmetric Planner Adaptation, a trajectory-based training paradigm that aligns the high-level planner with agent tools via backward reconstruction and forward refinement. To validate the performance of HTAA, we conduct experiments on a real-world internal dataset, InfoVerify, based on the POI validation workflow of China's largest online large-scale ride-hailing platform, featuring long-horizon executable tool trajectories. Experiments on InfoVerify and widely-used benchmarks show that HTAA consistently achieves higher task success rates, requires short tool calling trajectories, and significantly reduces context overhead compared to strong baselines. Furthermore, in a production deployment, HTAA substantially reduces manual validation effort and operational cost, demonstrating its practical efficacy.

Bing Zhao, Chenfei Wu, Deqing Li et al.

We present Qwen-Image-2.0, an omni-capable image generation foundation model that unifies high-fidelity generation and precise image editing within a single framework. Despite recent progress, existing models still struggle with ultra-long text rendering, multilingual typography, high-resolution photorealism, robust instruction following, and efficient deployment, especially in text-rich and compositionally complex scenarios. Qwen-Image-2.0 addresses these challenges by coupling Qwen3-VL as the condition encoder with a Multimodal Diffusion Transformer for joint condition-target modeling, supported by large-scale data curation and a customized multi-stage training pipeline. This enables strong multimodal understanding while preserving flexible generation and editing capabilities. The model supports instructions of up to 1K tokens for generating text-rich content such as slides, posters, infographics, and comics, while significantly improving multilingual text fidelity and typography. It also enhances photorealistic generation with richer details, more realistic textures, and coherent lighting, and follows complex prompts more reliably across diverse styles. Extensive human evaluations show that Qwen-Image-2.0 substantially outperforms previous Qwen-Image models in both generation and editing, marking a step toward more general, reliable, and practical image generation foundation models.

Junhao Xiao, Shun Feng, Zhiyu Wu et al.

Audio-Visual Speaker Detection (AVSD) hinges on modeling both individual temporal continuity and inter-personal social context. Existing coupled architectures struggle to reconcile these tasks in shared representation spaces due to conflicting inductive biases: temporal modeling favors low-frequency smoothness, while inter-personal interaction requires high-frequency discriminability. We propose D$^2$Stream, a decoupled dual-stream framework that explicitly isolates these functionalities into parallel, task-specific branches. Specifically, the Intra-speaker Temporal Continuity (ITC) stream captures longitudinal stability, whereas the Inter-personal Social Relation (ISR) stream models transversal social cues. Quantitative gradient analysis reveals an evolutionary divergence in update directions, stabilizing at 86.1°, which confirms the inherent task conflict and the effectiveness of our structural decoupling. D$^2$Stream breaks the long-standing performance plateau, achieving a state-of-the-art 95.6% mAP on AVA-ActiveSpeaker and superior generalization on Columbia ASD, all within a lightweight and efficient design.

Zhiyuan Ma, Xinyue Liang, Rongyuan Wu et al.

It is highly desirable to obtain a model that can generate high-quality 3D meshes from text prompts in just seconds. While recent attempts have adapted pre-trained text-to-image diffusion models, such as Stable Diffusion (SD), into generators of 3D representations (e.g., Triplane), they often suffer from poor quality due to the lack of sufficient high-quality 3D training data. Aiming at overcoming the data shortage, we propose a novel training scheme, termed as Progressive Rendering Distillation (PRD), eliminating the need for 3D ground-truths by distilling multi-view diffusion models and adapting SD into a native 3D generator. In each iteration of training, PRD uses the U-Net to progressively denoise the latent from random noise for a few steps, and in each step it decodes the denoised latent into 3D output. Multi-view diffusion models, including MVDream and RichDreamer, are used in joint with SD to distill text-consistent textures and geometries into the 3D outputs through score distillation. Since PRD supports training without 3D ground-truths, we can easily scale up the training data and improve generation quality for challenging text prompts with creative concepts. Meanwhile, PRD can accelerate the inference speed of the generation model in just a few steps. With PRD, we train a Triplane generator, namely TriplaneTurbo, which adds only $2.5\%$ trainable parameters to adapt SD for Triplane generation. TriplaneTurbo outperforms previous text-to-3D generators in both efficiency and quality. Specifically, it can produce high-quality 3D meshes in 1.2 seconds and generalize well for challenging text input. The code is available at https://github.com/theEricMa/TriplaneTurbo.

Jinghan Yu, Junhao Xiao, Chenyu Zhu et al.

Existing text-guided image editing methods primarily rely on end-to-end pixel-level inpainting paradigm. Despite its success in simple scenarios, this paradigm still significantly struggles with compositional editing tasks that require precise local control and complex multi-object spatial reasoning. This paradigm is severely limited by 1) the implicit coupling of planning and execution, 2) the lack of object-level control granularity, and 3) the reliance on unstructured, pixel-centric modeling. To address these limitations, we propose I2E, a novel "Decompose-then-Action" paradigm that revisits image editing as an actionable interaction process within a structured environment. I2E utilizes a Decomposer to transform unstructured images into discrete, manipulable object layers and then introduces a physics-aware Vision-Language-Action Agent to parse complex instructions into a series of atomic actions via Chain-of-Thought reasoning. Further, we also construct I2E-Bench, a benchmark designed for multi-instance spatial reasoning and high-precision editing. Experimental results on I2E-Bench and multiple public benchmarks demonstrate that I2E significantly outperforms state-of-the-art methods in handling complex compositional instructions, maintaining physical plausibility, and ensuring multi-turn editing stability.

Hanqing Wang, Mingyu Liu, Xiaoyu Chen et al.

3D affordance grounding aims to highlight the actionable regions on 3D objects, which is crucial for robotic manipulation. Previous research primarily focused on learning affordance knowledge from static cues such as language and images, which struggle to provide sufficient dynamic interaction context that can reveal temporal and causal cues. To alleviate this predicament, we collect a comprehensive video-based 3D affordance dataset, \textit{VIDA}, which contains 38K human-object-interaction videos covering 16 affordance types, 38 object categories, and 22K point clouds. Based on \textit{VIDA}, we propose a strong baseline: VideoAfford, which activates multimodal large language models with additional affordance segmentation capabilities, enabling both world knowledge reasoning and fine-grained affordance grounding within a unified framework. To enhance action understanding capability, we leverage a latent action encoder to extract dynamic interaction priors from HOI videos. Moreover, we introduce a \textit{spatial-aware} loss function to enable VideoAfford to obtain comprehensive 3D spatial knowledge. Extensive experimental evaluations demonstrate that our model significantly outperforms well-established methods and exhibits strong open-world generalization with affordance reasoning abilities. All datasets and code will be publicly released to advance research in this area.

Zhenya Huang, Jiayu Liu, Xin Lin et al.

Math word problem (MWP) serves as a fundamental research topic in artificial intelligence (AI) dating back to 1960s. This research aims to advance the reasoning abilities of AI by mirroring the human-like cognitive intelligence. The mainstream technological paradigm has evolved from the early rule-based methods, to deep learning models, and is rapidly advancing towards large language models. However, the field still lacks a systematic taxonomy for the MWP survey along with a discussion of current development trends. Therefore, in this paper, we aim to comprehensively review related research in MWP solving through the lens of human cognition, to demonstrate how recent AI models are advancing in simulating human cognitive abilities. Specifically, we summarize 5 crucial cognitive abilities for MWP solving, including Problem Understanding, Logical Organization, Associative Memory, Critical Thinking, and Knowledge Learning. Focused on these abilities, we review two mainstream MWP models in recent 10 years: neural network solvers, and LLM based solvers, and discuss the core human-like abilities they demonstrated in their intricate problem-solving process. Moreover, we rerun all the representative MWP solvers and supplement their performance on 5 mainstream benchmarks for a unified comparison. To the best of our knowledge, this survey first comprehensively analyzes the influential MWP research of the past decade from the perspective of human reasoning cognition and provides an integrative overall comparison across existing approaches. We hope it can inspire further research in AI reasoning. Our repository is released on https://github.com/Ljyustc/FoI-MWP.

Zhiyuan Ma, Liangliang Zhao, Biqing Qi et al.

The most advanced diffusion models have recently adopted increasingly deep stacked networks (e.g., U-Net or Transformer) to promote the generative emergence capabilities of vision generation models similar to large language models (LLMs). However, progressively deeper stacked networks will intuitively cause numerical propagation errors and reduce noisy prediction capabilities on generative data, which hinders massively deep scalable training of vision generation models. In this paper, we first uncover the nature that neural networks being able to effectively perform generative denoising lies in the fact that the intrinsic residual unit has consistent dynamic property with the input signal's reverse diffusion process, thus supporting excellent generative abilities. Afterwards, we stand on the shoulders of two common types of deep stacked networks to propose a unified and massively scalable Neural Residual Diffusion Models framework (Neural-RDM for short), which is a simple yet meaningful change to the common architecture of deep generative networks by introducing a series of learnable gated residual parameters that conform to the generative dynamics. Experimental results on various generative tasks show that the proposed neural residual models obtain state-of-the-art scores on image's and video's generative benchmarks. Rigorous theoretical proofs and extensive experiments also demonstrate the advantages of this simple gated residual mechanism consistent with dynamic modeling in improving the fidelity and consistency of generated content and supporting large-scale scalable training. Code is available at https://github.com/Anonymous/Neural-RDM.

Rongyuan Wu, Lingchen Sun, Zhiyuan Ma et al.

The pre-trained text-to-image diffusion models have been increasingly employed to tackle the real-world image super-resolution (Real-ISR) problem due to their powerful generative image priors. Most of the existing methods start from random noise to reconstruct the high-quality (HQ) image under the guidance of the given low-quality (LQ) image. While promising results have been achieved, such Real-ISR methods require multiple diffusion steps to reproduce the HQ image, increasing the computational cost. Meanwhile, the random noise introduces uncertainty in the output, which is unfriendly to image restoration tasks. To address these issues, we propose a one-step effective diffusion network, namely OSEDiff, for the Real-ISR problem. We argue that the LQ image contains rich information to restore its HQ counterpart, and hence the given LQ image can be directly taken as the starting point for diffusion, eliminating the uncertainty introduced by random noise sampling. We finetune the pre-trained diffusion network with trainable layers to adapt it to complex image degradations. To ensure that the one-step diffusion model could yield HQ Real-ISR output, we apply variational score distillation in the latent space to conduct KL-divergence regularization. As a result, our OSEDiff model can efficiently and effectively generate HQ images in just one diffusion step. Our experiments demonstrate that OSEDiff achieves comparable or even better Real-ISR results, in terms of both objective metrics and subjective evaluations, than previous diffusion model-based Real-ISR methods that require dozens or hundreds of steps. The source codes are released at https://github.com/cswry/OSEDiff.

Biqing Qi, Yang Luo, Junqi Gao et al.

Deep State Space Models (SSMs) have proven effective in numerous task scenarios but face significant security challenges due to Adversarial Perturbations (APs) in real-world deployments. Adversarial Training (AT) is a mainstream approach to enhancing Adversarial Robustness (AR) and has been validated on various traditional DNN architectures. However, its effectiveness in improving the AR of SSMs remains unclear. While many enhancements in SSM components, such as integrating Attention mechanisms and expanding to data-dependent SSM parameterizations, have brought significant gains in Standard Training (ST) settings, their potential benefits in AT remain unexplored. To investigate this, we evaluate existing structural variants of SSMs with AT to assess their AR performance. We observe that pure SSM structures struggle to benefit from AT, whereas incorporating Attention yields a markedly better trade-off between robustness and generalization for SSMs in AT compared to other components. Nonetheless, the integration of Attention also leads to Robust Overfitting (RO) issues. To understand these phenomena, we empirically and theoretically analyze the output error of SSMs under AP. We find that fixed-parameterized SSMs have output error bounds strictly related to their parameters, limiting their AT benefits, while input-dependent SSMs may face the problem of error explosion. Furthermore, we show that the Attention component effectively scales the output error of SSMs during training, enabling them to benefit more from AT, but at the cost of introducing RO due to its high model complexity. Inspired by this, we propose a simple and effective Adaptive Scaling (AdS) mechanism that brings AT performance close to Attention-integrated SSMs without introducing the issue of RO. Our code is available at https://github.com/Biqing-Qi/Exploring-Adversarial-Robustness-of-Deep-State-Space-Models.git.

Kaiyan Zhang, Sihang Zeng, Ermo Hua et al.

Large Language Models (LLMs) have demonstrated remarkable capabilities across various domains and are moving towards more specialized areas. Recent advanced proprietary models such as GPT-4 and Gemini have achieved significant advancements in biomedicine, which have also raised privacy and security challenges. The construction of specialized generalists hinges largely on high-quality datasets, enhanced by techniques like supervised fine-tuning and reinforcement learning from human or AI feedback, and direct preference optimization. However, these leading technologies (e.g., preference learning) are still significantly limited in the open source community due to the scarcity of specialized data. In this paper, we present the UltraMedical collections, which consist of high-quality manual and synthetic datasets in the biomedicine domain, featuring preference annotations across multiple advanced LLMs. By utilizing these datasets, we fine-tune a suite of specialized medical models based on Llama-3 series, demonstrating breathtaking capabilities across various medical benchmarks. Moreover, we develop powerful reward models skilled in biomedical and general reward benchmark, enhancing further online preference learning within the biomedical LLM community. Datasets and models are available at https://github.com/TsinghuaC3I/UltraMedical

Song Tang, Yan Yang, Zhiyuan Ma et al.

In the classic setting of unsupervised domain adaptation (UDA), the labeled source data are available in the training phase. However, in many real-world scenarios, owing to some reasons such as privacy protection and information security, the source data is inaccessible, and only a model trained on the source domain is available. This paper proposes a novel deep clustering method for this challenging task. Aiming at the dynamical clustering at feature-level, we introduce extra constraints hidden in the geometric structure between data to assist the process. Concretely, we propose a geometry-based constraint, named semantic consistency on the nearest neighborhood (SCNNH), and use it to encourage robust clustering. To reach this goal, we construct the nearest neighborhood for every target data and take it as the fundamental clustering unit by building our objective on the geometry. Also, we develop a more SCNNH-compliant structure with an additional semantic credibility constraint, named semantic hyper-nearest neighborhood (SHNNH). After that, we extend our method to this new geometry. Extensive experiments on three challenging UDA datasets indicate that our method achieves state-of-the-art results. The proposed method has significant improvement on all datasets (as we adopt SHNNH, the average accuracy increases by over 3.0% on the large-scaled dataset). Code is available at https://github.com/tntek/N2DCX.

Jiaming Li, Chenyu Zhu, Zhiyuan Ma et al.

Reinforcement learning (RL) has shown extraordinary potential in aligning diffusion models to downstream tasks, yet most of them still suffer from significant reward hacking, which degrades generative diversity and quality by inducing visual mode collapse and amplifying unreliable rewards. We identify the root cause as the mode-seeking nature of these methods, which maximize expected reward without effectively constraining probability distribution over acceptable trajectories, causing concentration on a few high-reward paths. In contrast, we propose Trajectory Matching Policy Optimization (TMPO), which replaces scalar reward maximization with trajectory-level reward distribution matching. Specifically, TMPO introduces a Softmax Trajectory Balance (Softmax-TB) objective to match the policy probabilities of K trajectories to a reward-induced Boltzmann distribution. We prove that this objective inherits the mode-covering property of forward KL divergence, preserving coverage over all acceptable trajectories while optimizing reward. To further reduce multi-trajectory training time on large-scale flow-matching models, TMPO incorporates Dynamic Stochastic Tree Sampling, where trajectories share denoising prefixes and branch at dynamically scheduled steps, reducing redundant computation while improving training effectiveness. Extensive results across diverse alignment tasks such as human preference, compositional generation and text rendering show that TMPO improves generative diversity over state-of-the-art methods by 9.1%, and achieves competitive performance in all downstream and efficiency metrics, attaining the optimal trade-off between reward and diversity.

Jinhao Li, Yuhao Sun, Zhiyuan Ma et al.

Recurrent spiking neural networks (RSNNs) are a promising substrate for energy-efficient control policies, but training them for high-dimensional, long-horizon reinforcement learning remains challenging. Population-based, gradient-free optimization circumvents backpropagation through non-differentiable spike dynamics by estimating gradients. However, with finite populations, high variance of these estimates can induce harmful and overly aggressive update steps. Inspired by trust-region methods in reinforcement learning that constrain policy updates in distribution space, we propose \textbf{Signal-Adaptive Trust Regions (SATR)}, a distributional update rule that constrains relative change by bounding KL divergence normalized by an estimated signal energy. SATR automatically expands the trust region under strong signals and contracts it when updates are noise-dominated. We instantiate SATR for Bernoulli connectivity distributions, which have shown strong empirical performance for RSNN optimization. Across a suite of high-dimensional continuous-control benchmarks, SATR improves stability under limited populations and reaches competitive returns against strong baselines including PPO-LSTM. In addition, to make SATR practical at scale, we introduce a bitset implementation for binary spiking and binary weights, substantially reducing wall-clock training time and enabling fast RSNN policy search.

Zhiyuan Ma, Xiangyu Zhu, Guojun Qi et al.

Speech-driven 3D facial animation is important for many multimedia applications. Recent work has shown promise in using either Diffusion models or Transformer architectures for this task. However, their mere aggregation does not lead to improved performance. We suspect this is due to a shortage of paired audio-4D data, which is crucial for the Transformer to effectively perform as a denoiser within the Diffusion framework. To tackle this issue, we present DiffSpeaker, a Transformer-based network equipped with novel biased conditional attention modules. These modules serve as substitutes for the traditional self/cross-attention in standard Transformers, incorporating thoughtfully designed biases that steer the attention mechanisms to concentrate on both the relevant task-specific and diffusion-related conditions. We also explore the trade-off between accurate lip synchronization and non-verbal facial expressions within the Diffusion paradigm. Experiments show our model not only achieves state-of-the-art performance on existing benchmarks, but also fast inference speed owing to its ability to generate facial motions in parallel.

Zhiyuan Ma, Guoli Jia, Bowen Zhou

With the great success of text-conditioned diffusion models in creative text-to-image generation, various text-driven image editing approaches have attracted the attentions of many researchers. However, previous works mainly focus on discreteness-sensitive instructions such as adding, removing or replacing specific objects, background elements or global styles (i.e., hard editing), while generally ignoring subject-binding but semantically fine-changing continuity-sensitive instructions such as actions, poses or adjectives, and so on (i.e., soft editing), which hampers generative AI from generating user-customized visual contents. To mitigate this predicament, we propose a spatio-temporal guided adaptive editing algorithm AdapEdit, which realizes adaptive image editing by introducing a soft-attention strategy to dynamically vary the guiding degree from the editing conditions to visual pixels from both temporal and spatial perspectives. Note our approach has a significant advantage in preserving model priors and does not require model training, fine-tuning, extra data, or optimization. We present our results over a wide variety of raw images and editing instructions, demonstrating competitive performance and showing it significantly outperforms the previous approaches.

Zhiyuan Ma, zhihuan yu, Jianjun Li et al.

As a class of fruitful approaches, diffusion probabilistic models (DPMs) have shown excellent advantages in high-resolution image reconstruction. On the other hand, masked autoencoders (MAEs), as popular self-supervised vision learners, have demonstrated simpler and more effective image reconstruction and transfer capabilities on downstream tasks. However, they all require extremely high training costs, either due to inherent high temporal-dependence (i.e., excessively long diffusion steps) or due to artificially low spatial-dependence (i.e., human-formulated high mask ratio, such as 0.75). To the end, this paper presents LMD, a faster image reconstruction framework with latent masking diffusion. First, we propose to project and reconstruct images in latent space through a pre-trained variational autoencoder, which is theoretically more efficient than in the pixel-based space. Then, we combine the advantages of MAEs and DPMs to design a progressive masking diffusion model, which gradually increases the masking proportion by three different schedulers and reconstructs the latent features from simple to difficult, without sequentially performing denoising diffusion as in DPMs or using fixed high masking ratio as in MAEs, so as to alleviate the high training time-consumption predicament. Our approach allows for learning high-capacity models and accelerate their training (by 3x or more) and barely reduces the original accuracy. Inference speed in downstream tasks also significantly outperforms the previous approaches.

Yukun Wang, Longguang Wang, Zhiyuan Ma et al.

Despite the typical inversion-then-editing paradigm using text-to-image (T2I) models has demonstrated promising results, directly extending it to text-to-video (T2V) models still suffers severe artifacts such as color flickering and content distortion. Consequently, current video editing methods primarily rely on T2I models, which inherently lack temporal-coherence generative ability, often resulting in inferior editing results. In this paper, we attribute the failure of the typical editing paradigm to: 1) Tightly Spatial-temporal Coupling. The vanilla pivotal-based inversion strategy struggles to disentangle spatial-temporal information in the video diffusion model; 2) Complicated Spatial-temporal Layout. The vanilla cross-attention control is deficient in preserving the unedited content. To address these limitations, we propose a spatial-temporal decoupled guidance (STDG) and multi-frame null-text optimization strategy to provide pivotal temporal cues for more precise pivotal inversion. Furthermore, we introduce a self-attention control strategy to maintain higher fidelity for precise partial content editing. Experimental results demonstrate that our method (termed VideoDirector) effectively harnesses the powerful temporal generation capabilities of T2V models, producing edited videos with state-of-the-art performance in accuracy, motion smoothness, realism, and fidelity to unedited content.

Xinwei Long, Zhiyuan Ma, Ermo Hua et al.

Retrieval-augmented generation (RAG) has emerged to address the knowledge-intensive visual question answering (VQA) task. Current methods mainly employ separate retrieval and generation modules to acquire external knowledge and generate answers, respectively. We propose ReAuSE, an alternative to the previous RAG model for the knowledge-based VQA task, which seamlessly integrates knowledge retriever into the generative multi-modal large language model, serving as a built-in search engine. Specifically, our model functions both as a generative retriever and an accurate answer generator. It not only helps retrieve documents from the knowledge base by producing identifiers for each document, but it also answers visual questions based on the retrieved documents. Furthermore, we propose a reinforced retrieval calibration module from relevance feedback to improve retrieval performance and align with the preferences for accurate answer generation. Extensive experiments on two representative OKVQA and A-OKVQA datasets demonstrate significant improvements ranging from 2.9\% to 9.6\% across all evaluation metrics when compared to strong baselines.

Zhiyuan Ma, Jianjun Li, Guohui Li et al.

With the flourishing of social media platforms, vision-language pre-training (VLP) recently has received great attention and many remarkable progresses have been achieved. The success of VLP largely benefits from the information complementation and enhancement between different modalities. However, most of recent studies focus on cross-modal contrastive learning (CMCL) to promote image-text alignment by pulling embeddings of positive sample pairs together while pushing those of negative pairs apart, which ignores the natural asymmetry property between different modalities and requires large-scale image-text corpus to achieve arduous progress. To mitigate this predicament, we propose CMAL, a Cross-Modal Associative Learning framework with anchor points detection and cross-modal associative learning for VLP. Specifically, we first respectively embed visual objects and textual tokens into separate hypersphere spaces to learn intra-modal hidden features, and then design a cross-modal associative prompt layer to perform anchor point masking and swap feature filling for constructing a hybrid cross-modal associative prompt. Afterwards, we exploit a unified semantic encoder to learn their cross-modal interactive features for context adaptation. Finally, we design an associative mapping classification layer to learn potential associative mappings between modalities at anchor points, within which we develop a fresh self-supervised associative mapping classification task to boost CMAL's performance. Experimental results verify the effectiveness of CMAL, showing that it achieves competitive performance against previous CMCL-based methods on four common downstream vision-and-language tasks, with significantly fewer corpus. Especially, CMAL obtains new state-of-the-art results on SNLI-VE and REC (testA).

Guanwen Feng, Haoran Cheng, Yunan Li et al.

Implementing fine-grained emotion control is crucial for emotion generation tasks because it enhances the expressive capability of the generative model, allowing it to accurately and comprehensively capture and express various nuanced emotional states, thereby improving the emotional quality and personalization of generated content. Generating fine-grained facial animations that accurately portray emotional expressions using only a portrait and an audio recording presents a challenge. In order to address this challenge, we propose a visual attribute-guided audio decoupler. This enables the obtention of content vectors solely related to the audio content, enhancing the stability of subsequent lip movement coefficient predictions. To achieve more precise emotional expression, we introduce a fine-grained emotion coefficient prediction module. Additionally, we propose an emotion intensity control method using a fine-grained emotion matrix. Through these, effective control over emotional expression in the generated videos and finer classification of emotion intensity are accomplished. Subsequently, a series of 3DMM coefficient generation networks are designed to predict 3D coefficients, followed by the utilization of a rendering network to generate the final video. Our experimental results demonstrate that our proposed method, EmoSpeaker, outperforms existing emotional talking face generation methods in terms of expression variation and lip synchronization. Project page: https://peterfanfan.github.io/EmoSpeaker/

Yixiao Chen, Zhiyuan Ma, Guoli Jia et al.

Autoregressive transformers have recently shown impressive image generation quality and efficiency on par with state-of-the-art diffusion models. Unlike diffusion architectures, autoregressive models can naturally incorporate arbitrary modalities into a single, unified token sequence--offering a concise solution for multi-conditional image generation tasks. In this work, we propose $\textbf{ContextAR}$, a flexible and effective framework for multi-conditional image generation. ContextAR embeds diverse conditions (e.g., canny edges, depth maps, poses) directly into the token sequence, preserving modality-specific semantics. To maintain spatial alignment while enhancing discrimination among different condition types, we introduce hybrid positional encodings that fuse Rotary Position Embedding with Learnable Positional Embedding. We design Conditional Context-aware Attention to reduces computational complexity while preserving effective intra-condition perception. Without any fine-tuning, ContextAR supports arbitrary combinations of conditions during inference time. Experimental results demonstrate the powerful controllability and versatility of our approach, and show that the competitive perpormance than diffusion-based multi-conditional control approaches the existing autoregressive baseline across diverse multi-condition driven scenarios. Project page: $\href{https://context-ar.github.io/}{https://context-ar.github.io/.}$

Jiadong Pan, Zhiyuan Ma, Kaiyan Zhang et al.

Diffusion models have recently demonstrated exceptional performance in image generation task. However, existing image generation methods still significantly suffer from the dilemma of image reasoning, especially in logic-centered image generation tasks. Inspired by the success of Chain of Thought (CoT) and Reinforcement Learning (RL) in LLMs, we propose SRRL, a self-reflective RL algorithm for diffusion models to achieve reasoning generation of logical images by performing reflection and iteration across generation trajectories. The intermediate samples in the denoising process carry noise, making accurate reward evaluation difficult. To address this challenge, SRRL treats the entire denoising trajectory as a CoT step with multi-round reflective denoising process and introduces condition guided forward process, which allows for reflective iteration between CoT steps. Through SRRL-based iterative diffusion training, we introduce image reasoning through CoT into generation tasks adhering to physical laws and unconventional physical phenomena for the first time. Notably, experimental results of case study exhibit that the superior performance of our SRRL algorithm even compared with GPT-4o. The project page is https://jadenpan0.github.io/srrl.github.io/.

Jiabao Wei, Zhiyuan Ma

Virtual Try-ON (VTON) aims to synthesis specific person images dressed in given garments, which recently receives numerous attention in online shopping scenarios. Currently, the core challenges of the VTON task mainly lie in the fine-grained semantic extraction (i.e.,deep semantics) of the given reference garments during depth estimation and effective texture preservation when the garments are synthesized and warped onto human body. To cope with these issues, we propose DH-VTON, a deep text-driven virtual try-on model featuring a special hybrid attention learning strategy and deep garment semantic preservation module. By standing on the shoulder of a well-built pre-trained paint-by-example (abbr. PBE) approach, we present our DH-VTON pipeline in this work. Specifically, to extract the deep semantics of the garments, we first introduce InternViT-6B as fine-grained feature learner, which can be trained to align with the large-scale intrinsic knowledge with deep text semantics (e.g.,"neckline" or "girdle") to make up for the deficiency of the commonly adopted CLIP encoder. Based on this, to enhance the customized dressing abilities, we further introduce Garment-Feature ControlNet Plus (abbr. GFC+) module and propose to leverage a fresh hybrid attention strategy for training, which can adaptively integrate fine-grained characteristics of the garments into the different layers of the VTON model, so as to achieve multi-scale features preservation effects. Extensive experiments on several representative datasets demonstrate that our method outperforms previous diffusion-based and GAN-based approaches, showing competitive performance in preserving garment details and generating authentic human images.

Yu Liu, Wenxiao Zhang, Cong Cao et al.

Retrieval-augmented generation (RAG) is widely used to ground Large Language Models (LLMs) for multi-hop question answering. Recent work mainly focused on improving answer accuracy via fine-tuning and structured or reinforcement-based optimization. However, reliable reasoning in response generation faces three challenges: 1) Reasoning Collapse. Reasoning in multi-hop QA is inherently complex due to multi-hop composition and is further destabilized by noisy retrieval. 2) Reasoning-answer inconsistency. Due to the intrinsic uncertainty of LLM generation and exposure to evidence--distractor mixtures, models may produce correct answers that are not faithfully supported by their intermediate reasoning or evidence. 3) Loss of format control. Traditional chain-of-thought generation often deviates from required structured output formats, leading to incomplete or malformed structured content. To address these challenges, we propose CRAFT (Calibrated Reasoning with Answer-Faithful Traces), a Group Relative Policy Optimization (GRPO) based reinforcement learning framework that trains models to perform faithful reasoning during response generation. CRAFT employs dual reward mechanisms to optimize multi-hop reasoning: deterministic rewards ensure structural correctness while judge-based rewards verify semantic faithfulness. This optimization framework supports controllable trace variants that enable systematic analysis of how structure and scale affect reasoning performance and faithfulness. Experiments on three multi-hop QA benchmarks show that CRAFT improves both answer accuracy and reasoning faithfulness across model scales, with the CRAFT 7B model achieving competitive performance with closed-source LLMs across multiple reasoning trace settings.

Yu Liu, Yanbing Liu, Fangfang Yuan et al.

Recent advances in large language models (LLMs) and dense retrievers have driven significant progress in retrieval-augmented generation (RAG). However, existing approaches face significant challenges in complex reasoning-oriented multi-hop retrieval tasks: 1) Ineffective reasoning-oriented planning: Prior methods struggle to generate robust multi-step plans for complex queries, as rule-based decomposers perform poorly on out-of-template questions. 2) Suboptimal reasoning-driven retrieval: Related methods employ limited query reformulation, leading to iterative retrieval loops that often fail to locate golden documents. 3) Insufficient reasoning-guided filtering: Prevailing methods lack the fine-grained reasoning to effectively filter salient information from noisy results, hindering utilization of retrieved knowledge. Fundamentally, these limitations all stem from the weak coupling between retrieval and reasoning in current RAG architectures. We introduce the Orchestrated Planner-Executor Reasoning Architecture (OPERA), a novel reasoning-driven retrieval framework. OPERA's Goal Planning Module (GPM) decomposes questions into sub-goals, which are executed by a Reason-Execute Module (REM) with specialized components for precise reasoning and effective retrieval. To train OPERA, we propose Multi-Agents Progressive Group Relative Policy Optimization (MAPGRPO), a novel variant of GRPO. Experiments on complex multi-hop benchmarks show OPERA's superior performance, validating both the MAPGRPO method and OPERA's design.