Bingxuan Li

Shengbin Yue, Wei Chen, Siyuan Wang et al.

We propose DISC-LawLLM, an intelligent legal system utilizing large language models (LLMs) to provide a wide range of legal services. We adopt legal syllogism prompting strategies to construct supervised fine-tuning datasets in the Chinese Judicial domain and fine-tune LLMs with legal reasoning capability. We augment LLMs with a retrieval module to enhance models' ability to access and utilize external legal knowledge. A comprehensive legal benchmark, DISC-Law-Eval, is presented to evaluate intelligent legal systems from both objective and subjective dimensions. Quantitative and qualitative results on DISC-Law-Eval demonstrate the effectiveness of our system in serving various users across diverse legal scenarios. The detailed resources are available at https://github.com/FudanDISC/DISC-LawLLM.

Wei Chen, Qiushi Wang, Zefei Long et al.

We propose Multiple Experts Fine-tuning Framework to build a financial large language model (LLM), DISC-FinLLM. Our methodology improves general LLMs by endowing them with multi-turn question answering abilities, domain text processing capabilities, mathematical computation skills, and retrieval-enhanced generation capabilities. We build a financial instruction-tuning dataset named DISC-FIN-SFT, including instruction samples of four categories (consulting, NLP tasks, computing and retrieval-augmented generation). Evaluations conducted on multiple benchmarks demonstrate that our model performs better than baseline models in various financial scenarios. Further resources can be found at https://github.com/FudanDISC/DISC-FinLLM.

Jiayu Liu, Cheng Qian, Zhenhailong Wang et al.

Planning for real-world problems by language models often involves both world and user constraints, which may not be fully specified upfront and are progressively disclosed through interaction. However, existing benchmarks still underexplore adaptive planning under such progressively revealed dual constraints. To address this gap, we introduce AdaPlanBench, a dynamic interactive benchmark for evaluating whether Large Language Model (LLM) agents can adaptively plan and re-plan under progressively revealed world and user constraints. AdaPlanBench is built on 307 household tasks, with a scalable constraint construction pipeline that augments each task with dual constraints. At runtime, agents interact with the environment in a multi-turn protocol where hidden constraints are revealed only when the agent proposes a plan that violates them, requiring iterative plan revision under accumulating feedback. This makes planning challenging, as agents must infer and track constraints from feedback while re-planning effectively. Experiments on ten leading LLMs show that adaptive planning under dual constraints remains challenging, with the best model reaching only 67.75% accuracy. We further observe that performance degrades as more constraints accumulate, with user constraints posing a particularly large challenge and failures often stemming from weaker physical grounding and reduced effectiveness. These results establish AdaPlanBench as a testbed for dual-constrained interactive planning and highlight the challenge of reliable adaptation to dynamically revealed constraints in LLM agents.

Jiateng Liu, Bingxuan Li, Zhenhailong Wang et al.

We dream of AI agents that can read arbitrary designs and construct real-world objects from reusable building blocks. As a first step toward this vision, we study whether multimodal large language models (MLLMs) possess the visual grounding and spatial reasoning capabilities required for brick assembly. We formulate brick assembly as a sequential decision-making problem, where each step involves two subtasks: brick selection, identifying the target brick from candidate components, and brick pose estimation, predicting where and how the selected brick should be placed. To support this study, we introduce BC-Bench (Brick Construction Benchmark), the first benchmark for evaluating MLLMs on assembly with diverse bricks. Experiments show that current state-of-the-art MLLMs remain far from reliable builders, struggling with fine-grained brick selection and failing at precise pose estimation. To bridge this gap, we propose Brick-Composer, a learning framework that equips MLLMs with assembly skills through three complementary signals: Human Design Sparks, which provide affordance-rich construction demonstrations; World Feedback, which grounds predicted actions in visual and physical consequences; and Synthetic Experience, which scales learning beyond existing object designs. Brick-Composer improves brick selection accuracy by over three times, substantially reduces pose estimation errors, and raises strict step-level assembly success from less than 1% to around 15%. After training, a Qwen-3-8B can correctly compose up to 42% of the steps for a complete object, suggesting that MLLMs can acquire assembly capabilities through targeted, physically grounded learning.

Yifeng Wang, Zhi Tu, Yiwen Xiang et al. · cmu

The success of Computer Vision (CV) relies heavily on manually annotated data. However, it is prohibitively expensive to annotate images in key domains such as healthcare, where data labeling requires significant domain expertise and cannot be easily delegated to crowd workers. To address this challenge, we propose a neuro-symbolic approach called Rapid, which infers image labeling rules from a small amount of labeled data provided by domain experts and automatically labels unannotated data using the rules. Specifically, Rapid combines pre-trained CV models and inductive logic learning to infer the logic-based labeling rules. Rapid achieves a labeling accuracy of 83.33% to 88.33% on four image labeling tasks with only 12 to 39 labeled samples. In particular, Rapid significantly outperforms finetuned CV models in two highly specialized tasks. These results demonstrate the effectiveness of Rapid in learning from small data and its capability to generalize among different tasks. Code and our dataset are publicly available at https://github.com/Neural-Symbolic-Image-Labeling/

Cheng Qian, Hyeonjeong Ha, Jiayu Liu et al.

Large multimodal models (LMMs) have rapidly advanced in perception and reasoning; however, it remains unclear whether these capabilities generalize to discovering visually grounded solutions in open-ended environments, beyond pattern recognition. In such settings, intelligence requires more than answering well-posed questions: it involves identifying how elements in a scene can be repurposed in non-obvious yet physically feasible ways. This form of creative problem-solving is central to human intelligence, but remains largely untested in current benchmarks. To evaluate this ability, we introduce MM-CreativityBench, a benchmark for affordance-grounded creative tool use in visually rich, physically constrained environments. Each instance presents a scenario image with structured views of candidate entities and their parts, enabling fine-grained, interactive evaluation of how models iteratively inspect the scene, identify relevant affordances, and compose visually and physically grounded solutions. Our experiments show that current LMMs often fall short, not due to lack of generative capability, but because they do not sustain grounded exploration. Models often overlook relevant entities, under-examine critical parts, or hallucinate attributes not grounded in the image. Motivated by this failure mode, we propose affordance-grounded alignment, which casts creative tool use as a preference learning problem. Using Direct Preference Optimization, we encourage models to prefer attribute-affordance reasoning grounded in visual evidence over hallucinated alternatives. In addition, we incorporate supervision derived from an affordance knowledge base to guide broader entity exploration and multi-turn planning. Our results show consistent gains in selecting the correct entities and parts, while substantially reducing hallucination and grounding-related errors.

Songyan Zhao, Bingxuan Li, Yufei Tian et al.

Automatic melody-to-lyric (M2L) generation aims to create lyrics that align with a given melody. While most previous approaches generate lyrics from scratch, revision, editing plain text draft to fit it into the melody, offers a much more flexible and practical alternative. This enables broad applications, such as generating lyrics from flexible inputs (keywords, themes, or full text that needs refining to be singable), song translation (preserving meaning across languages while keeping the melody intact), or style transfer (adapting lyrics to different genres). This paper introduces REFFLY (REvision Framework For LYrics), the first revision framework for editing and generating melody-aligned lyrics. We train the lyric revision module using our curated synthesized melody-aligned lyrics dataset, enabling it to transform plain text into lyrics that align with a given melody. To further enhance the revision ability, we propose training-free heuristics aimed at preserving both semantic meaning and musical consistency throughout the editing process. Experimental results demonstrate the effectiveness of REFFLY across various tasks (e.g. lyrics generation, song translation), showing that our model outperforms strong baselines, including Lyra (Tian et al., 2023) and GPT-4, by 25% in both musicality and text quality.

Yifan Shen, Jiateng Liu, Xinzhuo Li et al.

Generative world models have shown promise for simulating dynamic environments, yet egocentric video remains challenging due to rapid viewpoint changes, frequent hand-object interactions, and goal-directed procedures whose evolution depends on latent human intent. Existing approaches either focus on hand-centric instructional synthesis with limited scene evolution, perform static view translation without modeling action dynamics, or rely on dense supervision, such as camera trajectories, long video prefixes, synchronized multicamera capture, etc. In this work, we introduce EgoForge, an egocentric goal-directed world simulator that generates coherent, first-person video rollouts from minimal static inputs: a single egocentric image, a high-level instruction, and an optional auxiliary exocentric view. To improve intent alignment and temporal consistency, we propose VideoDiffusionNFT, a trajectory-level reward-guided refinement that optimizes goal completion, temporal causality, scene consistency, and perceptual fidelity during diffusion sampling. Extensive experiments show EgoForge achieves consistent gains in semantic alignment, geometric stability, and motion fidelity over strong baselines, and robust performance in real-world smart-glasses experiments.

Cheng Qian, Emre Can Acikgoz, Bingxuan Li et al.

Agents built on vision-language models increasingly face tasks that demand anticipating future states rather than relying on short-horizon reasoning. Generative world models offer a promising remedy: agents could use them as external simulators to foresee outcomes before acting. This paper empirically examines whether current agents can leverage such world models as tools to enhance their cognition. Across diverse agentic and visual question answering tasks, we observe that some agents rarely invoke simulation (fewer than 1%), frequently misuse predicted rollouts (approximately 15%), and often exhibit inconsistent or even degraded performance (up to 5%) when simulation is available or enforced. Attribution analysis further indicates that the primary bottleneck lies in the agents' capacity to decide when to simulate, how to interpret predicted outcomes, and how to integrate foresight into downstream reasoning. These findings underscore the need for mechanisms that foster calibrated, strategic interaction with world models, paving the way toward more reliable anticipatory cognition in future agent systems.

Bingxuan Li, Jeonghwan Kim, Cheng Qian et al.

Overlapping calendar invitations force busy professionals to repeatedly decide which meetings to attend, reschedule, or decline. We refer to this preference-driven decision process as calendar conflict resolution. Automating this decision process is crucial yet challenging. Scheduling logistics can drain hours, and human delegation often fails at scale, which motivates us to ask: Can we trust large language models (LLMs) or language agents to manage time? To enable a systematic study of this question, we introduce CalConflictBench, a benchmark for long-horizon calendar conflict resolution. In CalConflictBench, conflicts are presented to agents round-by-round over a calendar year, requiring them to infer and adapt to user preferences progressively. Our experiments show that current LLM agents perform poorly with high error rates, e.g., Qwen-3-30B-Think has an average error rate of 35%. To address this gap, we propose PEARL, a reinforcement-learning framework that (i) augments the language agent with an external preference memory that stores and updates inferred strategies (e.g., attendee priorities, topic importance, time/location preferences), and (ii) optimizes the agent with round-wise rewards that directly supervise decision correctness, ranking quality, and memory usage across rounds. Experiments on CalConflictBench show that PEARL achieves an error reduction rate of 0.76 and a 55% improvement in average error rate compared to the strongest baseline.

Yunxiang Zhang, Bingxuan Li, Alexandr Kuznetsov et al.

Neural image representations have emerged as a promising approach for encoding and rendering visual data. Combined with learning-based workflows, they demonstrate impressive trade-offs between visual fidelity and memory footprint. Existing methods in this domain, however, often rely on fixed data structures that suboptimally allocate memory or compute-intensive implicit models, hindering their practicality for real-time graphics applications. Inspired by recent advancements in radiance field rendering, we introduce Image-GS, a content-adaptive image representation based on 2D Gaussians. Leveraging a custom differentiable renderer, Image-GS reconstructs images by adaptively allocating and progressively optimizing a group of anisotropic, colored 2D Gaussians. It achieves a favorable balance between visual fidelity and memory efficiency across a variety of stylized images frequently seen in graphics workflows, especially for those showing non-uniformly distributed features and in low-bitrate regimes. Moreover, it supports hardware-friendly rapid random access for real-time usage, requiring only 0.3K MACs to decode a pixel. Through error-guided progressive optimization, Image-GS naturally constructs a smooth level-of-detail hierarchy. We demonstrate its versatility with several applications, including texture compression, semantics-aware compression, and joint image compression and restoration.

Xuying Ning, Katherine Tieu, Dongqi Fu et al.

Recent large language models (LLMs) have demonstrated strong capabilities in understanding and generating code, from competitive programming to repository-level software engineering. In emerging agentic systems, code is no longer only a target output. It increasingly serves as an operational substrate for agent reasoning, acting, environment modeling, and execution-based verification. We frame this shift through the lens of agent harnesses and introduce code as agent harness: a unified view that centers code as the basis for agent infrastructure. To systematically study this perspective, we organize the survey around three connected layers. First, we study the harness interface, where code connects agents to reasoning, action, and environment modeling. Second, we examine harness mechanisms: planning, memory, and tool use for long-horizon execution, together with feedback-driven control and optimization that make harness reliable and adaptive. Third, we discuss scaling the harness from single-agent systems to multi-agent settings, where shared code artifacts support multi-agent coordination, review, and verification. Across these layers, we summarize representative methods and practical applications of code as agent harness, spanning coding assistants, GUI/OS automation, embodied agents, scientific discovery, personalization and recommendation, DevOps, and enterprise workflows. We further outline open challenges for harness engineering, including evaluation beyond final task success, verification under incomplete feedback, regression-free harness improvement, consistent shared state across multiple agents, human oversight for safety-critical actions, and extensions to multimodal environments. By centering code as the harness of agentic AI, this survey provides a unified roadmap toward executable, verifiable, and stateful AI agent systems.

Bohui Shen, Wei Zhang, Xubiao Liu et al.

Positron emission tomography (PET) serves as an essential tool for diagnosis of encephalopathy and brain science research. However, it suffers from the limited choice of tracers. Nowadays, with the wide application of PET imaging in neuropsychiatric treatment, 6-18F-fluoro-3, 4-dihydroxy-L-phenylalanine (DOPA) has been found to be more effective than 18F-labeled fluorine-2-deoxyglucose (FDG) in the field. Nevertheless, due to the complexity of its preparation and other limitations, DOPA is far less widely used than FDG. To address this issue, a tracer conversion invertible neural network (TC-INN) for image projection is developed to map FDG images to DOPA images through deep learning. More diagnostic information is obtained by generating PET images from FDG to DOPA. Specifically, the proposed TC-INN consists of two separate phases, one for training traceable data, the other for rebuilding new data. The reference DOPA PET image is used as a learning target for the corresponding network during the training process of tracer conversion. Meanwhile, the invertible network iteratively estimates the resultant DOPA PET data and compares it to the reference DOPA PET data. Notably, the reversible model employs variable enhancement technique to achieve better power generation. Moreover, image registration needs to be performed before training due to the angular deviation of the acquired FDG and DOPA data information. Experimental results exhibited excellent generation capability in mapping between FDG and DOPA, suggesting that PET tracer conversion has great potential in the case of limited tracer applications.

Yu Guan, Bohui Shen, Xinchong Shi et al.

Attenuation correction (AC) is essential for the generation of artifact-free and quantitatively accurate positron emission tomography (PET) images. However, AC of PET faces challenges including inter-scan motion and erroneous transformation of structural voxel-intensities to PET attenuation-correction factors. Nowadays, the problem of AC for quantitative PET have been solved to a large extent after the commercial availability of devices combining PET with computed tomography (CT). Meanwhile, considering the feasibility of a deep learning approach for PET AC without anatomical imaging, this paper develops a PET AC method, which uses deep learning to generate continuously valued CT images from non-attenuation corrected PET images for AC on brain PET imaging. Specifically, an invertible network combined with the variable augmentation strategy that can achieve the bidirectional inference processes is proposed for synthetic CT generation (IVNAC). To evaluate the performance of the proposed algorithm, we conducted a comprehensive study on a total of 1440 data from 37 clinical patients using comparative algorithms (such as Cycle-GAN and Pix2pix). Perceptual analysis and quantitative evaluations illustrate that the invertible network for PET AC outperforms other existing AC models, which demonstrates the potential of the proposed method and the feasibility of achieving brain PET AC without CT.

Bin Huang, Xubiao Liu, Lei Fang et al.

Positron emission tomography (PET) is an advanced medical imaging technique that plays a crucial role in non-invasive clinical diagnosis. However, while reducing radiation exposure through low-dose PET scans is beneficial for patient safety, it often results in insufficient statistical data. This scarcity of data poses significant challenges for accurately reconstructing high-quality images, which are essential for reliable diagnostic outcomes. In this research, we propose a diffusion transformer model (DTM) guided by joint compact prior (JCP) to enhance the reconstruction quality of low-dose PET imaging. In light of current research findings, we present a pioneering PET reconstruction model that integrates diffusion and transformer models for joint optimization. This model combines the powerful distribution mapping abilities of diffusion models with the capacity of transformers to capture long-range dependencies, offering significant advantages for low-dose PET reconstruction. Additionally, the incorporation of the lesion refining block and penalized weighted least squares (PWLS) enhance the recovery capability of lesion regions and preserves detail information, solving blurring problems in lesion areas and texture details of most deep learning frameworks. Experimental results demonstrate the effectiveness of DTM in enhancing image quality and preserving critical clinical information for low-dose PET scans. Our approach not only reduces radiation exposure risks but also provides a more reliable PET imaging tool for early disease detection and patient management.

Dylan Zhang, Yanshan Lin, Zhengkun Wu et al.

Learning from past experience benefits from two complementary forms of memory: episodic traces -- raw trajectories of what happened -- and consolidated abstractions distilled across many episodes into reusable, schema-like lessons. Recent agentic-memory systems pursue the consolidated form: an LLM rewrites past trajectories into a textual memory bank that it continuously updates with new interactions, promising self-improving agents without parameter updates. Yet we find that such consolidated memories produced by today's LLMs are often faulty even when derived from useful experiences. As consolidation proceeds, memory utility first rises, then degrades, and can fall below the no-memory baseline. More surprisingly, even when consolidating from ground-truth solutions, GPT-5.4 fails on 54% of a set of ARC-AGI problems it had previously solved without memory. We trace the regression to the consolidation step rather than the underlying experience: the same trajectories yield qualitatively different memories under different update schedules, and an episodic-only control that simply retains those trajectories remains competitive with the consolidators we test. In a controlled ARC-AGI Stream environment that exposes Retain, Delete, and Consolidate actions, agents preserve raw episodes by default and double the accuracy of their forced-consolidation counterparts; disabling consolidation entirely (episodic management only) matches this auto regime. Practically, robust agent memory should treat raw episodes as first-class evidence and gate consolidation explicitly rather than firing it after every interaction. Looking forward, reliable agentic memory will require LLMs that can consolidate without overwriting the evidence they depend on.

Bingxuan Li, Yiming Cui, Yicheng He et al.

Sound effects build an essential layer of multimodal storytelling, shaping the emotional atmosphere and the narrative semantics of videos. Despite recent advancement in video-text-to-audio (VT2A), the current formulation faces three key limitations: First, an imbalance between visual and textual conditioning that leads to visual dominance; Second, the absence of a concrete definition for fine-grained controllable generation; Third, weak instruction understanding and following, as existing datasets rely on brief categorical tags. To address these limitations, we introduce EchoFoley, a new task designed for video-grounded sound generation with both event level local control and hierarchical semantic control. Our symbolic representation for sounding events specifies when, what, and how each sound is produced within a video or instruction, enabling fine-grained controls like sound generation, insertion, and editing. To support this task, we construct EchoFoley-6k, a large-scale, expert-curated benchmark containing over 6,000 video-instruction-annotation triplets. Building upon this foundation, we propose EchoVidia a sounding-event-centric agentic generation framework with slow-fast thinking strategy. Experiments show that EchoVidia surpasses recent VT2A models by 40.7% in controllability and 12.5% in perceptual quality.

Cheng Qian, Hyeonjeong Ha, Jiayu Liu et al.

Recent advances in large language models have led to strong performance on reasoning and environment-interaction tasks, yet their ability for creative problem-solving remains underexplored. We study this capability through the lens of creative tool use, where a model repurposes available objects by reasoning about their affordances and attributes rather than relying on canonical usage. As a first step, we introduce CreativityBench, a benchmark for evaluating affordance-based creativity in LLMs. To this end, we build a large-scale affordance knowledge base (KB) with 4K entities and 150K+ affordance annotations, explicitly linking objects, parts, attributes, and actionable uses. Building on this KB, we generate 14K grounded tasks that require identifying non-obvious yet physically plausible solutions under constraints. Evaluations across 10 state-of-the-art LLMs, including closed and open-source models, show that models can often select a plausible object, but fail to identify the correct parts, their affordances, and the underlying physical mechanism needed to solve the task, leading to a significant drop in performance. Furthermore, improvements from model scaling quickly saturate, strong general reasoning does not reliably translate to creative affordance discovery, and common inference-time strategies such as Chain-of-Thought yield limited gains. These results suggest that creative tool use remains a major challenge for current models, and that CreativityBench provides a useful testbed for studying this missing dimension of intelligence, with potential implications for planning and reasoning modules in future agents.

Yingqian Wu, Qiushi Wang, Zefei Long et al.

Financial report generation tasks range from macro- to micro-economics analysis, also requiring extensive data analysis. Existing LLM models are usually fine-tuned on simple QA tasks and cannot comprehensively analyze real financial scenarios. Given the complexity, financial companies often distribute tasks among departments. Inspired by this, we propose FinTeam, a financial multi-agent collaborative system, with a workflow with four LLM agents: document analyzer, analyst, accountant, and consultant. We train these agents with specific financial expertise using constructed datasets. We evaluate FinTeam on comprehensive financial tasks constructed from real online investment forums, including macroeconomic, industry, and company analysis. The human evaluation shows that by combining agents, the financial reports generate from FinTeam achieved a 62.00% acceptance rate, outperforming baseline models like GPT-4o and Xuanyuan. Additionally, FinTeam's agents demonstrate a 7.43% average improvement on FinCUGE and a 2.06% accuracy boost on FinEval. Project is available at https://github.com/FudanDISC/DISC-FinLLM/.

Kunlun Zhu, Zijia Liu, Bingxuan Li et al.

Large Language Model (LLM) agents, which integrate planning, memory, reflection, and tool-use modules, have shown promise in solving complex, multi-step tasks. Yet their sophisticated architectures amplify vulnerability to cascading failures, where a single root-cause error propagates through subsequent decisions, leading to task failure. Current systems lack a framework that can comprehensively understand agent error in a modular and systemic way, and therefore fail to detect these errors accordingly. We address this gap with three contributions. First, we introduce the AgentErrorTaxonomy, a modular classification of failure modes spanning memory, reflection, planning, action, and system-level operations. Second, we construct AgentErrorBench, the first dataset of systematically annotated failure trajectories from ALFWorld, GAIA, and WebShop, grounding error analysis in real-world agent rollouts. Third, we propose AgentDebug, a debugging framework that isolates root-cause failures and provides corrective feedback, enabling agents to recover and iteratively improve. Experiments on AgentErrorBench show that AgentDebug achieves 24% higher all-correct accuracy and 17% higher step accuracy compared to the strongest baseline. Beyond detection, the targeted feedback generated by AgentDebug enables LLM agents to iteratively recover from failures, yielding up to 26% relative improvements in task success across ALFWorld, GAIA, and WebShop. These results establish principled debugging as a pathway to more reliable and adaptive LLM agents. The code and data will be available at https://github.com/ulab-uiuc/AgentDebug

Fang Chen, Weifeng Zhang, Xingyu Ai et al.

Positron emission tomography (PET) is widely used to assess metabolic activity, but its application is limited by the availability of radiotracers. 18F-labeled fluorodeoxyglucose (18F-FDG) is the most commonly used tracer but shows limited effectiveness for certain tumors. In contrast, 6-18F-fluoro-3,4-dihydroxy-L-phenylalanine (18F-DOPA) offers higher specificity for neuroendocrine tumors and neurological disorders. However, the complexity of its synthesis process and constraints on transportation time have limited its clinical application. Among different forms of raw data acquired by the scanner, sinogram is a commonly used representation in PET imaging. Therefore, modeling in projection domain enables more direct utilization of the original information, potentially reducing the accumulation errors during the image reconstruction process. Inspired by these factors, this study proposes a prior-guided joint diffusion model (PJDM) for transforming 18F-FDG PET sinograms into 18F-DOPA PET sinograms. During inference, an initial synthetic 18F-DOPA PET sinogram is first generated using a higher-order hybrid sampler. This sinogram is then degraded and serves as an additional condition to guide the iterative refinement process. Experimental results demonstrated that PJDM effectively improved both sinogram quality and the final synthetic outcomes. The code is available at: https://github.com/yqx7150/PJDM.

Mengxiao Geng, Zijie Chen, Ran Hong et al.

Positron emission tomography (PET) reconstruction is a critical challenge in molecular imaging, often hampered by noise amplification, structural blurring, and detail loss due to sparse sampling and the ill-posed nature of inverse problems. The three-dimensional discrete Gaussian representation (GR), which efficiently encodes 3D scenes using parameterized discrete Gaussian distributions, has shown promise in computer vision. In this work, we pro-pose a novel GR-Diffusion framework that synergistically integrates the geometric priors of GR with the generative power of diffusion models for 3D low-dose whole-body PET reconstruction. GR-Diffusion employs GR to generate a reference 3D PET image from projection data, establishing a physically grounded and structurally explicit benchmark that overcomes the low-pass limitations of conventional point-based or voxel-based methods. This reference image serves as a dual guide during the diffusion process, ensuring both global consistency and local accuracy. Specifically, we employ a hierarchical guidance mechanism based on the GR reference. Fine-grained guidance leverages differences to refine local details, while coarse-grained guidance uses multi-scale difference maps to correct deviations. This strategy allows the diffusion model to sequentially integrate the strong geometric prior from GR and recover sub-voxel information. Experimental results on the UDPET and Clinical datasets with varying dose levels show that GR-Diffusion outperforms state-of-the-art methods in enhancing 3D whole-body PET image quality and preserving physiological details.

Masahiro Yoshida, Bingxuan Li, Songyan Zhao et al.

We propose CoLyricist, an AI-assisted lyric writing tool designed to support the typical workflows of experienced lyricists and enhance their creative efficiency. While lyricists have unique processes, many follow common stages. Tools that fail to accommodate these stages challenge integration into creative practices. Existing research and tools lack sufficient understanding of these songwriting stages and their associated challenges, resulting in ineffective designs. Through a formative study involving semi-structured interviews with 10 experienced lyricists, we identified four key stages: Theme Setting, Ideation, Drafting Lyrics, and Melody Fitting. CoLyricist addresses these needs by incorporating tailored AI-driven support for each stage, optimizing the lyric writing process to be more seamless and efficient. To examine whether this workflow-aligned design also benefits those without prior experience, we conducted a user study with 16 participants, including both experienced and novice lyricists. Results showed that CoLyricist enhances the songwriting experience across skill levels. Novice users especially appreciated the Melody-Fitting feature, while experienced users valued the Ideation support.

Jiateng Liu, Rushi Wang, Bingxuan Li et al.

Recent advances have enabled general computer-use agents that interpret screens and execute grounded actions from human instructions, yet they still struggle to generalize to unseen and evolving interfaces. While improving agent capability remains important, agent compatible interface design offers a complementary path by aligning interaction semantics with agent prior knowledge. In this paper, we revisit Nielsen 10 usability heuristics through the lens of computer-use agents, identifying which principles naturally transfer, where implicit design assumptions create agent specific failures, and how safe additive augmentations can improve robustness without harming human usability. To evaluate these ideas, we introduce UI-Verse, a suite of controlled environments built around functionally similar interfaces with different applied heuristics. Experiments show that our augmented heuristics consistently improve task completion and modestly improve efficiency, with combined heuristics yielding further gains. Human studies further show that these designs preserve the original interaction workflow without observable usability regressions. Overall, our findings highlight interface design as a practical complementary avenue for improving the reliability and generalization of computer use agents.

Xueqing Wu, Yuheng Ding, Bingxuan Li et al.

The ability of large vision-language models (LVLMs) to critique and correct their reasoning is an essential building block towards their self-improvement. However, a systematic analysis of such capabilities in LVLMs is still lacking. We propose VISCO, the first benchmark to extensively analyze the fine-grained critique and correction capabilities of LVLMs. Compared to existing work that uses a single scalar value to critique the entire reasoning [4], VISCO features dense and fine-grained critique, requiring LVLMs to evaluate the correctness of each step in the chain-of-thought and provide natural language explanations to support their judgments. Extensive evaluation of 24 LVLMs demonstrates that human-written critiques significantly enhance the performance after correction, showcasing the potential of the self-improvement strategy. However, the model-generated critiques are less helpful and sometimes detrimental to the performance, suggesting that critique is the crucial bottleneck. We identified three common patterns in critique failures: failure to critique visual perception, reluctance to "say no", and exaggerated assumption of error propagation. To address these issues, we propose an effective LookBack strategy that revisits the image to verify each piece of information in the initial reasoning. LookBack significantly improves critique and correction performance by up to 13.5%.

Bingxuan Li, Yiwei Wang, Jiuxiang Gu et al.

Chart generation aims to generate code to produce charts satisfying the desired visual properties, e.g., texts, layout, color, and type. It has great potential to empower the automatic professional report generation in financial analysis, research presentation, education, and healthcare. In this work, we build a vision-language model (VLM) based multi-agent framework for effective automatic chart generation. Generating high-quality charts requires both strong visual design skills and precise coding capabilities that embed the desired visual properties into code. Such a complex multi-modal reasoning process is difficult for direct prompting of VLMs. To resolve these challenges, we propose METAL, a multi-agent framework that decomposes the task of chart generation into the iterative collaboration among specialized agents. METAL achieves 5.2% improvement over the current best result in the chart generation task. The METAL framework exhibits the phenomenon of test-time scaling: its performance increases monotonically as the logarithmic computational budget grows from 512 to 8192 tokens. In addition, we find that separating different modalities during the critique process of METAL boosts the self-correction capability of VLMs in the multimodal context.

Bingxuan Li, Simo Du, Yue Guo

Clinical expertise improves not only by acquiring medical knowledge, but by accumulating experience that yields reusable diagnostic patterns. Recent LLMs-based diagnostic agents have shown promising progress in clinical reasoning for decision support. However, most approaches treat cases independently, limiting experience reuse and continual adaptation. We propose SEA, a self-learning diagnostic agent with cognitively inspired dual-memory module. We design a reinforcement training framework tailored to our designed agent for joint optimization of reasoning and memory management. We evaluate SEA in two complementary settings. On standard evaluation with MedCaseReasoning dataset, SEA achieves 92.46% accuracy, outperforming the strongest baseline by +19.6%, demonstrating the benefit of jointly optimizing reasoning and memory. On the long-horizon with ER-Reason dataset, SEA attains the best final accuracy (0.7214) and the largest improvement (+0.35 Acc@100), while baseline methods show limited or unstable gains. Expert evaluation further indicates that rules consolidated from SEA show strong clinical correctness, usefulness and trust, suggesting that the induced rules in dual-memory module are reliable and practically meaningful. Overall, SEA improves both diagnostic reasoning ability and continual learning by effectively transforming experience into reusable knowledge.

Yining Hong, Rui Sun, Bingxuan Li et al.

AI agents today are mostly siloed - they either retrieve and reason over vast amount of digital information and knowledge obtained online; or interact with the physical world through embodied perception, planning and action - but rarely both. This separation limits their ability to solve tasks that require integrated physical and digital intelligence, such as cooking from online recipes, navigating with dynamic map data, or interpreting real-world landmarks using web knowledge. We introduce Embodied Web Agents, a novel paradigm for AI agents that fluidly bridge embodiment and web-scale reasoning. To operationalize this concept, we first develop the Embodied Web Agents task environments, a unified simulation platform that tightly integrates realistic 3D indoor and outdoor environments with functional web interfaces. Building upon this platform, we construct and release the Embodied Web Agents Benchmark, which encompasses a diverse suite of tasks including cooking, navigation, shopping, tourism, and geolocation - all requiring coordinated reasoning across physical and digital realms for systematic assessment of cross-domain intelligence. Experimental results reveal significant performance gaps between state-of-the-art AI systems and human capabilities, establishing both challenges and opportunities at the intersection of embodied cognition and web-scale knowledge access. All datasets, codes and websites are publicly available at our project page https://embodied-web-agent.github.io/.

Jiateng Liu, Zhenhailong Wang, Rushi Wang et al.

General-purpose computer-use agents have shown impressive performance across diverse digital environments. However, our new benchmark, OSExpert-Eval, indicates they remain far less helpful than human experts. Although inference-time scaling enables adaptation, these agents complete complex tasks inefficiently with degraded performance, transfer poorly to unseen UIs, and struggle with fine-grained action sequences. To solve the problem, we introduce a GUI-based depth-first search (GUI-DFS) exploration algorithm to comprehensively explore and verify an environment's unit functions. The agent then exploits compositionality between unit skills to self-construct a curriculum for composite tasks. To support fine-grained actions, we curate a database of action primitives for agents to discover during exploration; these are saved as a skill set once the exploration is complete. We use the learned skills to improve the agent's performance and efficiency by (1) enriching agents with ready-to-use procedural knowledge, allowing them to plan only once for long trajectories and generate accurate actions, and (2) enabling them to end inference-time scaling earlier by realizing their boundary of capabilities. Extensive experiments show that our environment-learned agent takes a meaningful step toward expert-level computer use, achieving a around 20 percent performance gain on OSExpert-Eval and closing the efficiency gap to humans by around 80 percent

Shujian Jiao, Bingxuan Li, Lei Wang et al.

Proteins are essential to life's processes, underpinning evolution and diversity. Advances in sequencing technology have revealed millions of proteins, underscoring the need for sophisticated pre-trained protein models for biological analysis and AI development. Facebook's ESM2, the most advanced protein language model to date, leverages a masked prediction task for unsupervised learning, crafting amino acid representations with notable biochemical accuracy. Yet, it lacks in delivering functional protein insights, signaling an opportunity for enhancing representation quality.Our study addresses this gap by incorporating protein family classification into ESM2's training.This approach, augmented with Community Propagation-Based Clustering Algorithm, improves global protein representations, while a contextual prediction task fine-tunes local amino acid accuracy. Significantly, our model achieved state-of-the-art results in several downstream experiments, demonstrating the power of combining global and local methodologies to substantially boost protein representation quality.

Yu Zhou, Bingxuan Li, Mohan Tang et al. · meta-ai

Large multimodal models (LMMs) often struggle to recognize novel concepts, as they rely on pre-trained knowledge and have limited ability to capture subtle visual details. Domain-specific knowledge gaps in training also make them prone to confusing visually similar, commonly misrepresented, or low-resource concepts. To help LMMs better align nuanced visual features with language, improving their ability to recognize and reason about novel or rare concepts, we propose a Contrastive visual Data Augmentation (CoDA) strategy. CoDA extracts key contrastive textual and visual features of target concepts against the known concepts they are misrecognized as, and then uses multimodal generative models to produce targeted synthetic data. Automatic filtering of extracted features and augmented images is implemented to guarantee their quality, as verified by human annotators. We show the effectiveness and efficiency of CoDA on low-resource concept and diverse scene recognition datasets including INaturalist and SUN. We additionally collect NovelSpecies, a benchmark dataset consisting of newly discovered animal species that are guaranteed to be unseen by LMMs. LLaVA-1.6 1-shot updating results on these three datasets show CoDA significantly improves SOTA visual data augmentation strategies by 12.3% (NovelSpecies), 5.1% (SUN), and 6.0% (iNat) absolute gains in accuracy.

Ran Hong, Yuxia Huang, Lei Liu et al.

PET imaging is widely employed for observing biological metabolic activities within the human body. However, numerous benign conditions can cause increased uptake of radiopharmaceuticals, confounding differentiation from malignant tumors. Several studies have indicated that dual-time PET imaging holds promise in distinguishing between malignant and benign tumor processes. Nevertheless, the hour-long distribution period of radiopharmaceuticals post-injection complicates the determination of optimal timing for the second scan, presenting challenges in both practical applications and research. Notably, we have identified that delay time PET imaging can be framed as an image-to-image conversion problem. Motivated by this insight, we propose a novel spatial-temporal guided diffusion transformer probabilistic model (st-DTPM) to solve dual-time PET imaging prediction problem. Specifically, this architecture leverages the U-net framework that integrates patch-wise features of CNN and pixel-wise relevance of Transformer to obtain local and global information. And then employs a conditional DDPM model for image synthesis. Furthermore, on spatial condition, we concatenate early scan PET images and noisy PET images on every denoising step to guide the spatial distribution of denoising sampling. On temporal condition, we convert diffusion time steps and delay time to a universal time vector, then embed it to each layer of model architecture to further improve the accuracy of predictions. Experimental results demonstrated the superiority of our method over alternative approaches in preserving image quality and structural information, thereby affirming its efficacy in predictive task.

Bin Huang, Kang Chen, Bingxuan Li et al.

Building large-scale foundation model for PET imaging is hindered by limited access to labeled data and insufficient computational resources. To overcome data scarcity and efficiency limitations, we propose ALL-PET, a low-resource, low-shot PET foundation model operating directly in projection domain. ALL-PET leverages a latent diffusion model (LDM) with three key innovations. First, we design a Radon mask augmentation strategy (RMAS) that generates over 200,000 structurally diverse training samples by projecting randomized image-domain masks into sinogram space, significantly improving generalization with minimal data. This is extended by a dynamic multi-mask (DMM) mechanism that varies mask quantity and distribution, enhancing data diversity without added model complexity. Second, we implement positive/negative mask constraints to embed strict geometric consistency, reducing parameter burden while preserving generation quality. Third, we introduce transparent medical attention (TMA), a parameter-free, geometry-driven mechanism that enhances lesion-related regions in raw projection data. Lesion-focused attention maps are derived from coarse segmentation, covering both hypermetabolic and hypometabolic areas, and projected into sinogram space for physically consistent guidance. The system supports clinician-defined ROI adjustments, ensuring flexible, interpretable, and task-adaptive emphasis aligned with PET acquisition physics. Experimental results show that ALL-PET achieves high-quality sinogram generation using only 500 samples, with performance comparable to models trained on larger datasets. ALL-PET generalizes across tasks including low-dose reconstruction, attenuation correction, delayed-frame prediction, and tracer separation, operating efficiently with memory use under 24GB.

Mengxiao Geng, Ran Hong, Bingxuan Li et al.

Ultra-low-dose positron emission tomography (PET) reconstruction holds significant potential for reducing patient radiation exposure and shortening examination times. However, it may also lead to increased noise and reduced imaging detail, which could decrease the image quality. In this study, we present a Double-Constraint Diffusion Model (DCDM), which freezes the weights of a pre-trained diffusion model and injects a trainable double-constraint controller into the encoding architecture, greatly reducing the number of trainable parameters for ultra-low-dose PET reconstruction. Unlike full fine-tuning models, DCDM can adapt to different dose levels without retraining all model parameters, thereby improving reconstruction flexibility. Specifically, the two constraint modules, named the Nuclear Transformer Constraint (NTC) and the Encoding Nexus Constraint (ENC), serve to refine the pre-trained diffusion model. The NTC leverages the nuclear norm as an approximation for matrix rank minimization, integrates the low-rank property into the Transformer architecture, and enables efficient information extraction from low-dose images and conversion into compressed feature representations in the latent space. Subsequently, the ENC utilizes these compressed feature representations to encode and control the pre-trained diffusion model, ultimately obtaining reconstructed PET images in the pixel space. In clinical reconstruction, the compressed feature representations from NTC help select the most suitable ENC for efficient unknown low-dose PET reconstruction. Experiments conducted on the UDPET public dataset and the Clinical dataset demonstrated that DCDM outperforms state-of-the-art methods on known dose reduction factors (DRF) and generalizes well to unknown DRF scenarios, proving valuable even at ultra-low dose levels, such as 1% of the full dose.

Chengfei Wu, Ronald Seoh, Bingxuan Li et al.

Recent advances in large vision-language models have led to impressive performance in visual question answering and multimodal reasoning. However, it remains unclear whether these models genuinely perform grounded visual reasoning or rely on superficial patterns and dataset biases. In this work, we introduce MagiC, a comprehensive benchmark designed to evaluate grounded multimodal cognition, assessing not only answer accuracy but also the quality of step-by-step reasoning and its alignment with relevant visual evidence. Our benchmark includes approximately 5,500 weakly supervised QA examples generated from strong model outputs and 900 human-curated examples with fine-grained annotations, including answers, rationales, and bounding box groundings. We evaluate 15 vision-language models ranging from 7B to 70B parameters across four dimensions: final answer correctness, reasoning validity, grounding fidelity, and self-correction ability. MagiC further includes diagnostic settings to probe model robustness under adversarial visual cues and assess their capacity for introspective error correction. We introduce new metrics such as MagiScore and StepSense, and provide comprehensive analyses that reveal key limitations and opportunities in current approaches to grounded visual reasoning.

Bingxuan Li, Jiahao Wu, Yuan Xu et al.

Depth imaging is a foundational building block for broad applications, such as autonomous driving and virtual/augmented reality. Traditionally, depth cameras have relied on time-of-flight sensors or multi-lens systems to achieve physical depth measurements. However, these systems often face a trade-off between a bulky form factor and imprecise approximations, limiting their suitability for spatially constrained scenarios. Inspired by the emerging advancements of nano-optics, we present Nano-3D, a metasurface-based neural depth imaging solution with an ultra-compact footprint. Nano-3D integrates our custom-fabricated 700 nm thick TiO2 metasurface with a multi-module deep neural network to extract precise metric depth information from monocular metasurface-polarized imagery. We demonstrate the effectiveness of Nano-3D with both simulated and physical experiments. We hope the exhibited success paves the way for the community to bridge future graphics systems with emerging nanomaterial technologies through novel computational approaches.