Xinyu Wang

Yuliang Liu, Jiaxin Zhang, Dezhi Peng et al.

End-to-end scene text spotting has made significant progress due to its intrinsic synergy between text detection and recognition. Previous methods commonly regard manual annotations such as horizontal rectangles, rotated rectangles, quadrangles, and polygons as a prerequisite, which are much more expensive than using single-point. Our new framework, SPTS v2, allows us to train high-performing text-spotting models using a single-point annotation. SPTS v2 reserves the advantage of the auto-regressive Transformer with an Instance Assignment Decoder (IAD) through sequentially predicting the center points of all text instances inside the same predicting sequence, while with a Parallel Recognition Decoder (PRD) for text recognition in parallel, which significantly reduces the requirement of the length of the sequence. These two decoders share the same parameters and are interactively connected with a simple but effective information transmission process to pass the gradient and information. Comprehensive experiments on various existing benchmark datasets demonstrate the SPTS v2 can outperform previous state-of-the-art single-point text spotters with fewer parameters while achieving 19$\times$ faster inference speed. Within the context of our SPTS v2 framework, our experiments suggest a potential preference for single-point representation in scene text spotting when compared to other representations. Such an attempt provides a significant opportunity for scene text spotting applications beyond the realms of existing paradigms. Code is available at: https://github.com/Yuliang-Liu/SPTSv2.

Amir Rasouli, Randy Goebel, Matthew E. Taylor et al. · gatech, nvidia

Driving SMARTS is a regular competition designed to tackle problems caused by the distribution shift in dynamic interaction contexts that are prevalent in real-world autonomous driving (AD). The proposed competition supports methodologically diverse solutions, such as reinforcement learning (RL) and offline learning methods, trained on a combination of naturalistic AD data and open-source simulation platform SMARTS. The two-track structure allows focusing on different aspects of the distribution shift. Track 1 is open to any method and will give ML researchers with different backgrounds an opportunity to solve a real-world autonomous driving challenge. Track 2 is designed for strictly offline learning methods. Therefore, direct comparisons can be made between different methods with the aim to identify new promising research directions. The proposed setup consists of 1) realistic traffic generated using real-world data and micro simulators to ensure fidelity of the scenarios, 2) framework accommodating diverse methods for solving the problem, and 3) baseline method. As such it provides a unique opportunity for the principled investigation into various aspects of autonomous vehicle deployment.

Tongya Zheng, Xinchao Wang, Zunlei Feng et al.

Temporal graphs exhibit dynamic interactions between nodes over continuous time, whose topologies evolve with time elapsing. The whole temporal neighborhood of nodes reveals the varying preferences of nodes. However, previous works usually generate dynamic representation with limited neighbors for simplicity, which results in both inferior performance and high latency of online inference. Therefore, in this paper, we propose a novel method of temporal graph convolution with the whole neighborhood, namely Temporal Aggregation and Propagation Graph Neural Networks (TAP-GNN). Specifically, we firstly analyze the computational complexity of the dynamic representation problem by unfolding the temporal graph in a message-passing paradigm. The expensive complexity motivates us to design the AP (aggregation and propagation) block, which significantly reduces the repeated computation of historical neighbors. The final TAP-GNN supports online inference in the graph stream scenario, which incorporates the temporal information into node embeddings with a temporal activation function and a projection layer besides several AP blocks. Experimental results on various real-life temporal networks show that our proposed TAP-GNN outperforms existing temporal graph methods by a large margin in terms of both predictive performance and online inference latency. Our code is available at \url{https://github.com/doujiang-zheng/TAP-GNN}.

Hengwei Zhao, Xinyu Wang, Jingtao Li et al.

Positive-unlabeled learning (PU learning) in hyperspectral remote sensing imagery (HSI) is aimed at learning a binary classifier from positive and unlabeled data, which has broad prospects in various earth vision applications. However, when PU learning meets limited labeled HSI, the unlabeled data may dominate the optimization process, which makes the neural networks overfit the unlabeled data. In this paper, a Taylor variational loss is proposed for HSI PU learning, which reduces the weight of the gradient of the unlabeled data by Taylor series expansion to enable the network to find a balance between overfitting and underfitting. In addition, the self-calibrated optimization strategy is designed to stabilize the training process. Experiments on 7 benchmark datasets (21 tasks in total) validate the effectiveness of the proposed method. Code is at: https://github.com/Hengwei-Zhao96/T-HOneCls.

Xinyu Wang, Mingze Li, Sicheng Lyu et al.

Vision-Language-Action (VLA) models unify perception, reasoning, and control within a single policy, yet their multi-billion-parameter backbones and diffusion-based action heads make on-device deployment prohibitively expensive. Prior quantization efforts offer only partial solutions, compressing the LLM backbone while leaving the DiT action head at full precision, or resorting to mixed-precision schemes, driven by the belief that uniformly quantizing the action head is inherently unstable. We challenge this assumption with Omega-QVLA, the first training-free post-training quantization framework that compresses both the language backbone and the entire diffusion action head of a VLA model to a uniform W4A4 precision, eliminating the need for mixed-precision allocation. Omega-QVLA combines a composite SVD-Hadamard rotation that equalizes per-channel weight energy while diffusing residual activation outliers with per-step DiT activation scaling quantization that absorbs dynamic-range drift across denoising steps. On LIBERO, Omega-QVLA compresses Pi 0.5 and GR00T N1.5 to W4A4 with 98.0% and 87.8% task success rates, matching or exceeding their FP16 references of 97.1% and 87.0%, while reducing the static memory footprint by 71.3%. Real-world manipulation experiments further confirm smooth, accurate manipulation where prior methods fail. Code is available at https://github.com/UCMP13753/Omega-QVLA.

Xinyu Wang, Hanwei Wu, Zhenghan Tai et al.

Medication recommendation predicts medications for patient visits, but existing methods still face two key challenges. At the model level, traditional drug recommendation methods only predict structured drug codes with limited evidence grounding, while LLM agents can use richer clinical context but may lack safety verification and traceability. At the task level, existing benchmarks often use broad medication categories, which ignore subgroup-level safety differences and can lead to risk overestimation. We introduce the first fine-grained medication recommendation setting based on fourth-level ATC code generation. We propose Safe Prescription Agent (SafeRx-Agent), a knowledge-grounded multi-agent framework that uses patient context, external clinical knowledge, and safety verification to recommend traceable medication sets. Experimental results on MIMIC-III and MIMIC-IV datasets show that SafeRx-Agent improves fine-grained medication prediction accuracy while controlling drug interactions, contraindications, and medication set size.

Xinyu Wang, Ziyu Zhao, Ke Bai et al.

Data-aware post-training quantization (PTQ) minimizes a per-token reconstruction loss on a small calibration corpus, implicitly weighting positions by their empirical frequency. For \textbf{A}utomatic \textbf{S}peech \textbf{R}ecognition (ASR), this misaligns with tail-sensitive risk: names, numerals, and domain-specific words receive proportionally little calibration mass. We propose \textbf{Tail-Aware Reconstruction Quantization} (\TARQ), a label-free PTQ framework that shifts calibration toward the lexical tail via \textbf{\rareBAL}, a closed-form per-Linear-layer rule equalizing common/tail mass, paired with a metric-consistent residual correction. \TARQ\ requires no entity labels, no curated calibration set, no validation decoding, and no additional training. Across eight ASR backbones and six datasets at W4G128, \TARQ\ improves mean rare-\textbf{W}ord \textbf{E}rror \textbf{R}ate (rare-WER) without an aggregate-WER regression, achieves the lowest cross-corpus rare-WER swing among compared methods, and transfers to entity-rich benchmarks (ProfASR, ContextASR-Speech-En) without entity supervision.

Kevin Pu, Jim Yang, Angel Yuan et al. · utoronto

Knowledge workers frequently encounter repetitive web data entry tasks, like updating records or placing orders. Web automation increases productivity, but translating tasks to web actions accurately and extending to new specifications is challenging. Existing tools can automate tasks that perform the same logical trace of UI actions (e.g., input text in each field in order), but do not support tasks requiring different executions based on varied input conditions. We present DiLogics, a programming-by-demonstration system that utilizes NLP to assist users in creating web automation programs that handle diverse specifications. DiLogics first semantically segments input data to structured task steps. By recording user demonstrations for each step, DiLogics generalizes the web macros to novel but semantically similar task requirements. Our evaluation showed that non-experts can effectively use DiLogics to create automation programs that fulfill diverse input instructions. DiLogics provides an efficient, intuitive, and expressive method for developing web automation programs satisfying diverse specifications.

Xinyu Wang, Bohan Zhuang, Qi Wu

Large Language Models (LLMs), primarily trained on text-based datasets, exhibit exceptional proficiencies in understanding and executing complex linguistic instructions via text outputs. However, they falter when requests to generate non-text ones. Concurrently, modality conversion models, such as text-to-image, despite generating high-quality images, suffer from a lack of extensive textual pretraining. As a result, these models are only capable of accommodating specific image descriptions rather than comprehending more complex instructions. To bridge this gap, we propose a novel approach, \methodname, from a modality conversion perspective that evolves a text-based LLM into a multi-modal one. We specifically employ a minimal dataset to instruct LLMs to recognize the intended output modality as directed by the instructions. Consequently, the adapted LLM can effectively summon various off-the-shelf modality conversion models from the model zoos to generate non-text responses. This circumvents the necessity for complicated pretraining that typically requires immense quantities of paired multi-modal data, while simultaneously inheriting the extensive knowledge of LLMs and the ability of high-quality generative models. To evaluate and compare the adapted multi-modal LLM with its traditional counterparts, we have constructed a multi-modal instruction benchmark that solicits diverse modality outputs. The experiment results reveal that, with minimal training, LLMs can be conveniently adapted to comprehend requests for non-text responses, thus achieving higher flexibility in multi-modal scenarios. Code and data will be made available at https://github.com/xinke-wang/SwitchGPT.

Xinyu Wang, Jiong Cai, Yong Jiang et al.

Multi-modal named entity recognition (NER) and relation extraction (RE) aim to leverage relevant image information to improve the performance of NER and RE. Most existing efforts largely focused on directly extracting potentially useful information from images (such as pixel-level features, identified objects, and associated captions). However, such extraction processes may not be knowledge aware, resulting in information that may not be highly relevant. In this paper, we propose a novel Multi-modal Retrieval based framework (MoRe). MoRe contains a text retrieval module and an image-based retrieval module, which retrieve related knowledge of the input text and image in the knowledge corpus respectively. Next, the retrieval results are sent to the textual and visual models respectively for predictions. Finally, a Mixture of Experts (MoE) module combines the predictions from the two models to make the final decision. Our experiments show that both our textual model and visual model can achieve state-of-the-art performance on four multi-modal NER datasets and one multi-modal RE dataset. With MoE, the model performance can be further improved and our analysis demonstrates the benefits of integrating both textual and visual cues for such tasks.

Jingtao Li, Xinyu Wang, Shaoyu Wang et al.

Hyperspectral anomaly detection (HAD) involves identifying the targets that deviate spectrally from their surroundings, without prior knowledge. Recently, deep learning based methods have become the mainstream HAD methods, due to their powerful spatial-spectral feature extraction ability. However, the current deep detection models are optimized to complete a proxy task (two-step paradigm), such as background reconstruction or generation, rather than achieving anomaly detection directly. This leads to suboptimal results and poor transferability, which means that the deep model is trained and tested on the same image. In this paper, an unsupervised transferred direct detection (TDD) model is proposed, which is optimized directly for the anomaly detection task (one-step paradigm) and has transferability. Specially, the TDD model is optimized to identify the spectral deviation relationship according to the anomaly definition. Compared to learning the specific background distribution as most models do, the spectral deviation relationship is universal for different images and guarantees the model transferability. To train the TDD model in an unsupervised manner, an anomaly sample simulation strategy is proposed to generate numerous pairs of anomaly samples. Furthermore, a global self-attention module and a local self-attention module are designed to help the model focus on the "spectrally deviating" relationship. The TDD model was validated on four public HAD datasets. The results show that the proposed TDD model can successfully overcome the limitation of traditional model training and testing on a single image, and the model has a powerful detection ability and excellent transferability.

Xinyu Wang, Yongliang Shen, Jiong Cai et al.

The MultiCoNER shared task aims at detecting semantically ambiguous and complex named entities in short and low-context settings for multiple languages. The lack of contexts makes the recognition of ambiguous named entities challenging. To alleviate this issue, our team DAMO-NLP proposes a knowledge-based system, where we build a multilingual knowledge base based on Wikipedia to provide related context information to the named entity recognition (NER) model. Given an input sentence, our system effectively retrieves related contexts from the knowledge base. The original input sentences are then augmented with such context information, allowing significantly better contextualized token representations to be captured. Our system wins 10 out of 13 tracks in the MultiCoNER shared task.

Jingtao Li, Xinyu Wang, Hengwei Zhao et al.

Remote sensing anomaly detector can find the objects deviating from the background as potential targets for Earth monitoring. Given the diversity in earth anomaly types, designing a transferring model with cross-modality detection ability should be cost-effective and flexible to new earth observation sources and anomaly types. However, the current anomaly detectors aim to learn the certain background distribution, the trained model cannot be transferred to unseen images. Inspired by the fact that the deviation metric for score ranking is consistent and independent from the image distribution, this study exploits the learning target conversion from the varying background distribution to the consistent deviation metric. We theoretically prove that the large-margin condition in labeled samples ensures the transferring ability of learned deviation metric. To satisfy this condition, two large margin losses for pixel-level and feature-level deviation ranking are proposed respectively. Since the real anomalies are difficult to acquire, anomaly simulation strategies are designed to compute the model loss. With the large-margin learning for deviation metric, the trained model achieves cross-modality detection ability in five modalities including hyperspectral, visible light, synthetic aperture radar (SAR), infrared and low-light in zero-shot manner.

Jingtao Li, Xinyu Wang, Hengwei Zhao et al.

Anomaly segmentation in high spatial resolution (HSR) remote sensing imagery is aimed at segmenting anomaly patterns of the earth deviating from normal patterns, which plays an important role in various Earth vision applications. However, it is a challenging task due to the complex distribution and the irregular shapes of objects, and the lack of abnormal samples. To tackle these problems, an anomaly segmentation model based on pixel descriptors (ASD) is proposed for anomaly segmentation in HSR imagery. Specifically, deep one-class classification is introduced for anomaly segmentation in the feature space with discriminative pixel descriptors. The ASD model incorporates the data argument for generating virtual ab-normal samples, which can force the pixel descriptors to be compact for normal data and meanwhile to be diverse to avoid the model collapse problems when only positive samples participated in the training. In addition, the ASD introduced a multi-level and multi-scale feature extraction strategy for learning the low-level and semantic information to make the pixel descriptors feature-rich. The proposed ASD model was validated using four HSR datasets and compared with the recent state-of-the-art models, showing its potential value in Earth vision applications.

Hengwei Zhao, Yanfei Zhong, Xinyu Wang et al.

Hyperspectral imagery (HSI) one-class classification is aimed at identifying a single target class from the HSI by using only knowing positive data, which can significantly reduce the requirements for annotation. However, when one-class classification meets HSI, it is difficult for classifiers to find a balance between the overfitting and underfitting of positive data due to the problems of distribution overlap and distribution imbalance. Although deep learning-based methods are currently the mainstream to overcome distribution overlap in HSI multiclassification, few studies focus on deep learning-based HSI one-class classification. In this article, a weakly supervised deep HSI one-class classifier, namely, HOneCls, is proposed, where a risk estimator,the one-class risk estimator, is particularly introduced to make the fully convolutional neural network (FCN) with the ability of one class classification in the case of distribution imbalance. Extensive experiments (20 tasks in total) were conducted to demonstrate the superiority of the proposed classifier.

Zhengbang Zhu, Shenyu Zhang, Yuzheng Zhuang et al.

High-quality traffic flow generation is the core module in building simulators for autonomous driving. However, the majority of available simulators are incapable of replicating traffic patterns that accurately reflect the various features of real-world data while also simulating human-like reactive responses to the tested autopilot driving strategies. Taking one step forward to addressing such a problem, we propose Realistic Interactive TrAffic flow (RITA) as an integrated component of existing driving simulators to provide high-quality traffic flow for the evaluation and optimization of the tested driving strategies. RITA is developed with consideration of three key features, i.e., fidelity, diversity, and controllability, and consists of two core modules called RITABackend and RITAKit. RITABackend is built to support vehicle-wise control and provide traffic generation models from real-world datasets, while RITAKit is developed with easy-to-use interfaces for controllable traffic generation via RITABackend. We demonstrate RITA's capacity to create diversified and high-fidelity traffic simulations in several highly interactive highway scenarios. The experimental findings demonstrate that our produced RITA traffic flows exhibit all three key features, hence enhancing the completeness of driving strategy evaluation. Moreover, we showcase the possibility for further improvement of baseline strategies through online fine-tuning with RITA traffic flows.

Xinyu Wang, Changzhi Sun, Lian Cheng et al.

Verifiers are crucial components for enhancing modern LLMs' reasoning capability. Typicalverifiers require resource-intensive superviseddataset construction, which is costly and faceslimitations in data diversity. In this paper, wepropose LOVER, an unsupervised verifier regularized by logical rules. LOVER treats theverifier as a binary latent variable, utilizinginternal activations and enforcing three logical constraints on multiple reasoning paths:negation consistency, intra-group consistency,and inter-group consistency (grouped by thefinal answer). By incorporating logical rulesas priors, LOVER can leverage unlabeled examples and is directly compatible with any offthe-shelf LLMs. Experiments on 10 datasetsdemonstrate that LOVER significantly outperforms unsupervised baselines, achieving performance comparable to the supervised verifier(reaching its 95% level on average). The sourcecode is publicly available at https://github.com/wangxinyufighting/llm-lover.

Zhengliang Liu, Hanqi Jiang, Tianyang Zhong et al.

In this paper, we present a large-scale evaluation probing GPT-4V's capabilities and limitations for biomedical image analysis. GPT-4V represents a breakthrough in artificial general intelligence (AGI) for computer vision, with applications in the biomedical domain. We assess GPT-4V's performance across 16 medical imaging categories, including radiology, oncology, ophthalmology, pathology, and more. Tasks include modality recognition, anatomy localization, disease diagnosis, report generation, and lesion detection. The extensive experiments provide insights into GPT-4V's strengths and weaknesses. Results show GPT-4V's proficiency in modality and anatomy recognition but difficulty with disease diagnosis and localization. GPT-4V excels at diagnostic report generation, indicating strong image captioning skills. While promising for biomedical imaging AI, GPT-4V requires further enhancement and validation before clinical deployment. We emphasize responsible development and testing for trustworthy integration of biomedical AGI. This rigorous evaluation of GPT-4V on diverse medical images advances understanding of multimodal large language models (LLMs) and guides future work toward impactful healthcare applications.

Sitao Luan, Chenqing Hua, Qincheng Lu et al.

Homophily principle, \ie{} nodes with the same labels or similar attributes are more likely to be connected, has been commonly believed to be the main reason for the superiority of Graph Neural Networks (GNNs) over traditional Neural Networks (NNs) on graph-structured data, especially on node-level tasks. However, recent work has identified a non-trivial set of datasets where GNN's performance compared to the NN's is not satisfactory. Heterophily, i.e. low homophily, has been considered the main cause of this empirical observation. People have begun to revisit and re-evaluate most existing graph models, including graph transformer and its variants, in the heterophily scenario across various kinds of graphs, e.g. heterogeneous graphs, temporal graphs and hypergraphs. Moreover, numerous graph-related applications are found to be closely related to the heterophily problem. In the past few years, considerable effort has been devoted to studying and addressing the heterophily issue. In this survey, we provide a comprehensive review of the latest progress on heterophilic graph learning, including an extensive summary of benchmark datasets and evaluation of homophily metrics on synthetic graphs, meticulous classification of the most updated supervised and unsupervised learning methods, thorough digestion of the theoretical analysis on homophily/heterophily, and broad exploration of the heterophily-related applications. Notably, through detailed experiments, we are the first to categorize benchmark heterophilic datasets into three sub-categories: malignant, benign and ambiguous heterophily. Malignant and ambiguous datasets are identified as the real challenging datasets to test the effectiveness of new models on the heterophily challenge. Finally, we propose several challenges and future directions for heterophilic graph representation learning.

Tung Sum Thomas Kwok, Zeyong Zhang, Xinyu Wang et al.

Multi-step LLM reasoning over structured tables fails because planning and execution share no explicit cell-grounding contract. Existing methods constrain the planner to a left-to-right factorization at odds with table permutation invariance, and score intermediate states by generated content alone, overlooking cell grounding. We conduct a pilot study showing that diffusion language models (DLMs) produce more human-aligned and permutation-stable cell attention on tables than autoregressive models, with a 40.2% median reduction in attention-AUROC variability under row reordering. Motivated by this, we propose TABALIGN, a planned table reasoning framework that operationalizes the contract. TABALIGN pairs a masked DLM planner, whose bidirectional denoising emits plan steps as binary cell masks, with TABATTN, a lightweight verifier trained on 1,600 human-verified attention standards to score each step by its attention overlap with the plan-designated mask. Across eight benchmarks covering table question answering and fact verification, TABALIGN improves average accuracy by 15.76 percentage points over the strongest open-source baseline at comparable 8B-class scale, with a matched-backbone ablation attributing 2.87 percentage points of this gain to the DLM planner over an AR planner on a fixed reasoner. Cleaner DLM plans also accelerate downstream reasoning execution by 44.64%.

Yunlong Lyu, Yixuan Tang, Peng Chen et al.

Modern AI-integrated IDEs are shifting from passive code completion to proactive Next Edit Suggestions (NES). Unlike traditional autocompletion, NES is designed to construct a richer context from both recent user interactions and the broader codebase to suggest multi-line, cross-line, or even cross-file modifications. This evolution significantly streamlines the programming workflow into a tab-by-tab interaction and enhances developer productivity. Consequently, NES introduces a more complex context retrieval mechanism and sophisticated interaction patterns. However, existing studies focus almost exclusively on the security implications of standalone LLM-based code generation, ignoring the potential attack vectors posed by NES in modern AI-integrated IDEs. The underlying mechanisms of NES remain under-explored, and their security implications are not yet fully understood. In this paper, we conduct the first systematic security study of NES systems. First, we perform an in-depth dissection of the NES mechanisms to understand the newly introduced threat vectors. It is found that NES retrieves a significantly expanded context, including inputs from imperceptible user actions and global codebase retrieval, which increases the attack surfaces. Second, we conduct a comprehensive in-lab study to evaluate the security implications of NES. The evaluation results reveal that NES is susceptible to context poisoning and is sensitive to transactional edits and human-IDE interactions. Third, we perform a large-scale online survey involving over 200 professional developers to assess the perceptions of NES security risks in real-world development workflows. The survey results indicate a general lack of awareness regarding the potential security pitfalls associated with NES, highlighting the need for increased education and improved security countermeasures in AI-integrated IDEs.

Haifeng Wang, Hua Wu, Tian Wu et al.

In this report, we introduce ERNIE 5.0, a natively autoregressive foundation model desinged for unified multimodal understanding and generation across text, image, video, and audio. All modalities are trained from scratch under a unified next-group-of-tokens prediction objective, based on an ultra-sparse mixture-of-experts (MoE) architecture with modality-agnostic expert routing. To address practical challenges in large-scale deployment under diverse resource constraints, ERNIE 5.0 adopts a novel elastic training paradigm. Within a single pre-training run, the model learns a family of sub-models with varying depths, expert capacities, and routing sparsity, enabling flexible trade-offs among performance, model size, and inference latency in memory- or time-constrained scenarios. Moreover, we systematically address the challenges of scaling reinforcement learning to unified foundation models, thereby guaranteeing efficient and stable post-training under ultra-sparse MoE architectures and diverse multimodal settings. Extensive experiments demonstrate that ERNIE 5.0 achieves strong and balanced performance across multiple modalities. To the best of our knowledge, among publicly disclosed models, ERNIE 5.0 represents the first production-scale realization of a trillion-parameter unified autoregressive model that supports both multimodal understanding and generation. To facilitate further research, we present detailed visualizations of modality-agnostic expert routing in the unified model, alongside comprehensive empirical analysis of elastic training, aiming to offer profound insights to the community.

Yao Zhu, Xinyu Wang, Hong-Shuo Chen et al.

A novel learning solution to image steganalysis based on the green learning paradigm, called Green Steganalyzer (GS), is proposed in this work. GS consists of three modules: 1) pixel-based anomaly prediction, 2) embedding location detection, and 3) decision fusion for image-level detection. In the first module, GS decomposes an image into patches, adopts Saab transforms for feature extraction, and conducts self-supervised learning to predict an anomaly score of their center pixel. In the second module, GS analyzes the anomaly scores of a pixel and its neighborhood to find pixels of higher embedding probabilities. In the third module, GS focuses on pixels of higher embedding probabilities and fuses their anomaly scores to make final image-level classification. Compared with state-of-the-art deep-learning models, GS achieves comparable detection performance against S-UNIWARD, WOW and HILL steganography schemes with significantly lower computational complexity and a smaller model size, making it attractive for mobile/edge applications. Furthermore, GS is mathematically transparent because of its modular design.

Zhe Liu, Jinghua Hou, Xinyu Wang et al.

The benefit of transformers in large-scale 3D point cloud perception tasks, such as 3D object detection, is limited by their quadratic computation cost when modeling long-range relationships. In contrast, linear RNNs have low computational complexity and are suitable for long-range modeling. Toward this goal, we propose a simple and effective window-based framework built on LInear grOup RNN (i.e., perform linear RNN for grouped features) for accurate 3D object detection, called LION. The key property is to allow sufficient feature interaction in a much larger group than transformer-based methods. However, effectively applying linear group RNN to 3D object detection in highly sparse point clouds is not trivial due to its limitation in handling spatial modeling. To tackle this problem, we simply introduce a 3D spatial feature descriptor and integrate it into the linear group RNN operators to enhance their spatial features rather than blindly increasing the number of scanning orders for voxel features. To further address the challenge in highly sparse point clouds, we propose a 3D voxel generation strategy to densify foreground features thanks to linear group RNN as a natural property of auto-regressive models. Extensive experiments verify the effectiveness of the proposed components and the generalization of our LION on different linear group RNN operators including Mamba, RWKV, and RetNet. Furthermore, it is worth mentioning that our LION-Mamba achieves state-of-the-art on Waymo, nuScenes, Argoverse V2, and ONCE dataset. Last but not least, our method supports kinds of advanced linear RNN operators (e.g., RetNet, RWKV, Mamba, xLSTM and TTT) on small but popular KITTI dataset for a quick experience with our linear RNN-based framework.

Yue Ma, Xu Ye, Qinghe Wang et al.

Generating high-fidelity visual effects (VFX) typically demands massive datasets and prohibitive computational power due to the intricate coupling of spatial textures and temporal dynamics. In this paper, we introduce EasyVFX, a resource-efficient framework that achieves realistic VFX synthesis under stringent constraints. Our core philosophy lies in frequency-domain decomposition: we observe that the complexity of VFX can be significantly mitigated by decoupling high-frequency components, which represent intricate spatial appearances, from low-frequency components that encapsulate global motion dynamics. This spectral disentanglement transforms a high-dimensional learning problem into manageable sub-tasks, thereby lowering the optimization barrier and reducing data dependency. Building upon this insight, we propose a two-stage training paradigm. First, we design a Frequency-aware Mixture-of-Experts (Freq-MoE) architecture. By utilizing a soft routing mechanism, our model assigns specialized experts to distinct spectral bands, enabling them to cultivate robust priors for appearance and motion dynamics. This specialization allows the model to acquire foundational VFX knowledge with fewer GPU resources. Second, we introduce a Test-Time Training strategy powered by a novel Frequency-constraint Loss. This allows the pre-trained model to swiftly adapt to specific, unseen effects through localized optimizations, requiring only about 100 steps on a single GPU. Experimental results demonstrate that EasyVFX produces structurally consistent and visually stunning effects, proving that frequency-aware learning is a key catalyst for democratizing professional-grade VFX.

Tung Sum Thomas Kwok, Xinyu Wang, Hengzhi He et al.

A major challenge in training TableQA agents, compared to standard text- and image-based agents, is that answers cannot be inferred from a static input but must be reasoned through stepwise transformations of the table state, introducing multi-step reasoning complexity and environmental interaction. This leads to a research question: Can explicit feedback on table transformation action improve model reasoning capability? In this work, we introduce RE-Tab, a plug-and-play framework that architecturally enhances trajectory search via lightweight, training-free reward modeling by formulating the problem as a Partially Observable Markov Decision Process. We demonstrate that providing explicit verifiable rewards during State Transition (``What is the best action?'') and Simulative Reasoning (``Am I sure about the output?'') is crucial to steer the agent's navigation in table states. By enforcing stepwise reasoning with reward feedback in table transformations, RE-Tab achieves state-of-the-art performance in TableQA with almost 25\% drop in inference cost. Furthermore, a direct plug-and-play implementation of RE-Tab brings up to 41.77% improvement in QA accuracy and 33.33% drop in test-time inference samples for consistent answer. Consistent improvement pattern across various LLMs and state-of-the-art benchmarks further confirms RE-Tab's generalisability. The repository is available at https://github.com/ThomasK1018/RE_Tab .

Rui Min, Zile Qiao, Ze Xu et al.

Foundation agents have rapidly advanced in their ability to reason and interact with real environments, making the evaluation of their core capabilities increasingly important. While many benchmarks have been developed to assess agent performance, most concentrate on academic settings or artificially designed scenarios while overlooking the challenges that arise in real applications. To address this issue, we focus on a highly practical real-world setting, the e-commerce domain, which involves a large volume of diverse user interactions, dynamic market conditions, and tasks directly tied to real decision-making processes. To this end, we introduce EcomBench, a holistic E-commerce Benchmark designed to evaluate agent performance in realistic e-commerce environments. EcomBench is built from genuine user demands embedded in leading global e-commerce ecosystems and is carefully curated and annotated through human experts to ensure clarity, accuracy, and domain relevance. It covers multiple task categories within e-commerce scenarios and defines three difficulty levels that evaluate agents on key capabilities such as deep information retrieval, multi-step reasoning, and cross-source knowledge integration. By grounding evaluation in real e-commerce contexts, EcomBench provides a rigorous and dynamic testbed for measuring the practical capabilities of agents in modern e-commerce.

Xinyu Wang, Boyu Han, Zhenghan Tai et al.

Electronic Design Automation (EDA) tools such as KiCad offer powerful functionalities but remain difficult to use, particularly for beginners, due to their steep learning curves and fragmented documentation. To address this challenge, we present SmartonAI, an AI-assisted interaction system that integrates large language models into the EDA workflow, enabling natural language communication, intelligent task decomposition, and contextual plugin execution. SmartonAI consists of two main components: a Chat Plugin that breaks down user instructions into subtasks and retrieves tailored documentation, and a OneCommandLine Plugin that recommends and executes relevant plugins based on user intent. The system supports multilingual interaction and adapts to user feedback through incremental learning. Preliminary results suggest that SmartonAI significantly reduces onboarding time and enhances productivity, representing a promising step toward generalizable AI-assisted interaction paradigms for complex software systems.

Shihao Cai, Runnan Fang, Jialong Wu et al.

Conducting reinforcement learning (RL) in simulated environments offers a cost-effective and highly scalable way to enhance language-based agents. However, previous work has been limited to semi-automated environment synthesis or tasks lacking sufficient difficulty, offering little breadth or depth. In addition, the instability of simulated users integrated into these environments, along with the heterogeneity across simulated environments, poses further challenges for agentic RL. In this work, we propose: (1) a unified pipeline for automated and scalable synthesis of simulated environments associated with high-difficulty but easily verifiable tasks; and (2) an environment level RL algorithm that not only effectively mitigates user instability but also performs advantage estimation at the environment level, thereby improving training efficiency and stability. Comprehensive evaluations on agentic benchmarks, including tau-bench, tau2-Bench, and VitaBench, validate the effectiveness of our proposed method. Further in-depth analyses underscore its out-of-domain generalization.

Wenjun Peng, Xinyu Wang, Qi Wu

Large language models (LLMs) have revolutionized automated code generation, yet the evaluation of their real-world effectiveness remains limited by static benchmarks and simplistic metrics. We present ProxyWar, a novel framework that systematically assesses code generation quality by embedding LLM-generated agents within diverse, competitive game environments. Unlike existing approaches, ProxyWar evaluates not only functional correctness but also the operational characteristics of generated programs, combining automated testing, iterative code repair, and multi-agent tournaments to provide a holistic view of program behavior. Applied to a range of state-of-the-art coders and games, our approach uncovers notable discrepancies between benchmark scores and actual performance in dynamic settings, revealing overlooked limitations and opportunities for improvement. These findings highlight the need for richer, competition-based evaluation of code generation. Looking forward, ProxyWar lays a foundation for research into LLM-driven algorithm discovery, adaptive problem solving, and the study of practical efficiency and robustness, including the potential for models to outperform hand-crafted agents. The project is available at https://github.com/xinke-wang/ProxyWar.

Kuan Li, Zhongwang Zhang, Huifeng Yin et al.

Transcending human cognitive limitations represents a critical frontier in LLM training. Proprietary agentic systems like DeepResearch have demonstrated superhuman capabilities on extremely complex information-seeking benchmarks such as BrowseComp, a feat previously unattainable. We posit that their success hinges on a sophisticated reasoning pattern absent in open-source models: the ability to systematically reduce extreme uncertainty when navigating vast information landscapes. Based on this insight, we introduce WebSailor, a complete post-training methodology designed to instill this crucial capability. Our approach involves generating novel, high-uncertainty tasks through structured sampling and information obfuscation, RFT cold start, and an efficient agentic RL training algorithm, Duplicating Sampling Policy Optimization (DUPO). With this integrated pipeline, WebSailor significantly outperforms all opensource agents in complex information-seeking tasks, matching proprietary agents' performance and closing the capability gap.

Rui Dong, Qingyue Wu, Danny Ding et al.

Abstract semantics has proven to be instrumental for accelerating search-based program synthesis, by enabling the sound pruning of a set of incorrect programs (without enumerating them). One may expect faster synthesis with increasingly finer-grained abstract semantics. Unfortunately, to the best of our knowledge, this is not the case, yet. The reason is because, as abstraction granularity increases -- while fewer programs are enumerated -- pruning becomes more costly. This imposes a fundamental limit on the overall synthesis performance, which we aim to address in this work. Our key idea is to introduce an offline presynthesis phase, which consists of two steps. Given a DSL with abstract semantics, the first semantics modeling step constructs a tree automaton A for a space of inputs -- such that, for any program P and for any considered input I, A has a run that corresponds to P's execution on I under abstract semantics. Then, the second step builds an oracle O for A. This O enables fast pruning during synthesis, by allowing us to efficiently find exactly those DSL programs that satisfy a given input-output example under abstract semantics. We have implemented this presynthesis-based synthesis paradigm in a framework, Foresighter. On top of it, we have developed three instantiations for SQL, string transformation, and matrix manipulation. All of them significantly outperform prior work in the respective domains.

Fangda Ye, Yuxin Hu, Pengxiang Zhu et al.

Recent progress in deep research systems has been impressive, but evaluation still lags behind real user needs. Existing benchmarks predominantly assess final reports using fixed rubrics, failing to evaluate the underlying research process. Most also offer limited multimodal coverage, rely on synthetic tasks that do not reflect real-world query complexity, and cannot be refreshed as knowledge evolves. To address these gaps, we introduce MiroEval, a benchmark and evaluation framework for deep research systems. The benchmark comprises 100 tasks (70 text-only, 30 multimodal), all grounded in real user needs and constructed via a dual-path pipeline that supports periodic updates, enabling a live and evolving setting. The proposed evaluation suite assesses deep research systems along three complementary dimensions: adaptive synthesis quality evaluation with task-specific rubrics, agentic factuality verification via active retrieval and reasoning over both web sources and multimodal attachments, and process-centric evaluation audits how the system searches, reasons, and refines throughout its investigation. Evaluation across 13 systems yields three principal findings: the three evaluation dimensions capture complementary aspects of system capability, with each revealing distinct strengths and weaknesses across systems; process quality serves as a reliable predictor of overall outcome while revealing weaknesses invisible to output-level metrics; and multimodal tasks pose substantially greater challenges, with most systems declining by 3 to 10 points. The MiroThinker series achieves the most balanced performance, with MiroThinker-H1 ranking the highest overall in both settings. Human verification and robustness results confirm the reliability of the benchmark and evaluation framework. MiroEval provides a holistic diagnostic tool for the next generation of deep research agents.

Zhaopeng Feng, Liangcai Su, Zhen Zhang et al.

As large language models (LLMs) evolve into autonomous agents for long-horizon information-seeking, managing finite context capacity has become a critical bottleneck. Existing context management methods typically commit to a single fixed strategy throughout the entire trajectory. Such static designs may work well in some states, but they cannot adapt as the usefulness and reliability of the accumulated context evolve during long-horizon search. To formalize this challenge, we introduce a probabilistic framework that characterizes long-horizon success through two complementary dimensions: search efficiency and terminal precision. Building on this perspective, we propose AgentSwing, a state-aware adaptive parallel context management routing framework. At each trigger point, AgentSwing expands multiple context-managed branches in parallel and uses lookahead routing to select the most promising continuation. Experiments across diverse benchmarks and agent backbones show that AgentSwing consistently outperforms strong static context management methods, often matching or exceeding their performance with up to $3\times$ fewer interaction turns while also improving the ultimate performance ceiling of long-horizon web agents. Beyond the empirical gains, the proposed probabilistic framework provides a principled lens for analyzing and designing future context management strategies for long-horizon agents.

Yijing Yang, Vasileios Magoulianitis, Xinyu Wang et al.

A statistical attention localization (SAL) method is proposed to facilitate the object classification task in this work. SAL consists of three steps: 1) preliminary attention window selection via decision statistics, 2) attention map refinement, and 3) rectangular attention region finalization. SAL computes soft-decision scores of local squared windows and uses them to identify salient regions in Step 1. To accommodate object of various sizes and shapes, SAL refines the preliminary result and obtain an attention map of more flexible shape in Step 2. Finally, SAL yields a rectangular attention region using the refined attention map and bounding box regularization in Step 3. As an application, we adopt E-PixelHop, which is an object classification solution based on successive subspace learning (SSL), as the baseline. We apply SAL so as to obtain a cropped-out and resized attention region as an alternative input. Classification results of the whole image as well as the attention region are ensembled to achieve the highest classification accuracy. Experiments on the CIFAR-10 dataset are given to demonstrate the advantage of the SAL-assisted object classification method.

Zhengwei Tao, Jialong Wu, Wenbiao Yin et al.

The advent of Large Language Model (LLM)-powered agents has revolutionized artificial intelligence by enabling solutions to complex, open-ended tasks through web-based information-seeking (IS) capabilities. The scarcity of high-quality training data has limited the development of IS agents. Existing approaches typically adopt an information-driven paradigm that first collects web data and then generates questions based on the retrieval. However, this may lead to inconsistency between information structure and reasoning structure, question and answer. To mitigate, we propose a formalization-driven IS data synthesis framework WebShaper to construct a dataset. WebShaper systematically formalizes IS tasks through set theory. Central to the formalization is the concept of Knowledge Projections (KP), which enables precise control over reasoning structure by KP operation compositions. During synthesis, we begin by creating seed tasks, then use a multi-step expansion process. At each step, an agentic Expander expands the current formal question more complex with retrieval and validation tools based on our formalization. We train our model on the synthesized dataset. Experiment results demonstrate that WebShaper achieves state-of-the-art performance among open-sourced IS agents on GAIA and WebWalkerQA benchmarks.

Yangning Li, Yinghui Li, Xinyu Wang et al.

Multimodal Retrieval Augmented Generation (mRAG) plays an important role in mitigating the "hallucination" issue inherent in multimodal large language models (MLLMs). Although promising, existing heuristic mRAGs typically predefined fixed retrieval processes, which causes two issues: (1) Non-adaptive Retrieval Queries. (2) Overloaded Retrieval Queries. However, these flaws cannot be adequately reflected by current knowledge-seeking visual question answering (VQA) datasets, since the most required knowledge can be readily obtained with a standard two-step retrieval. To bridge the dataset gap, we first construct Dyn-VQA dataset, consisting of three types of "dynamic" questions, which require complex knowledge retrieval strategies variable in query, tool, and time: (1) Questions with rapidly changing answers. (2) Questions requiring multi-modal knowledge. (3) Multi-hop questions. Experiments on Dyn-VQA reveal that existing heuristic mRAGs struggle to provide sufficient and precisely relevant knowledge for dynamic questions due to their rigid retrieval processes. Hence, we further propose the first self-adaptive planning agent for multimodal retrieval, OmniSearch. The underlying idea is to emulate the human behavior in question solution which dynamically decomposes complex multimodal questions into sub-question chains with retrieval action. Extensive experiments prove the effectiveness of our OmniSearch, also provide direction for advancing mRAG. The code and dataset will be open-sourced at https://github.com/Alibaba-NLP/OmniSearch.

Peixin Zhang, Jun Sun, Mingtian Tan et al.

In recent years, the security issues of artificial intelligence have become increasingly prominent due to the rapid development of deep learning research and applications. Backdoor attack is an attack targeting the vulnerability of deep learning models, where hidden backdoors are activated by triggers embedded by the attacker, thereby outputting malicious predictions that may not align with the intended output for a given input. In this work, we propose a novel black-box backdoor attack based on machine unlearning. The attacker first augments the training set with carefully designed samples, including poison and mitigation data, to train a `benign' model. Then, the attacker posts unlearning requests for the mitigation samples to remove the impact of relevant data on the model, gradually activating the hidden backdoor. Since backdoors are implanted during the iterative unlearning process, it significantly increases the computational overhead of existing defense methods for backdoor detection or mitigation. To address this new security threat, we proposes two methods for detecting or mitigating such malicious unlearning requests. We conduct the experiment in both exact unlearning and approximate unlearning (i.e., SISA) settings. Experimental results indicate that: 1) our attack approach can successfully implant backdoor into the model, and sharding increases the difficult of attack; 2) our detection algorithms are effective in identifying the mitigation samples, while sharding reduces the effectiveness of our detection algorithms.

Sitao Luan, Qincheng Lu, Chenqing Hua et al.

Over the past decade, Graph Neural Networks (GNNs) have achieved great success on machine learning tasks with relational data. However, recent studies have found that heterophily can cause significant performance degradation of GNNs, especially on node-level tasks. Numerous heterophilic benchmark datasets have been put forward to validate the efficacy of heterophily-specific GNNs, and various homophily metrics have been designed to help recognize these challenging datasets. Nevertheless, there still exist multiple pitfalls that severely hinder the proper evaluation of new models and metrics: 1) lack of hyperparameter tuning; 2) insufficient evaluation on the truly challenging heterophilic datasets; 3) missing quantitative evaluation for homophily metrics on synthetic graphs. To overcome these challenges, we first train and fine-tune baseline models on $27$ most widely used benchmark datasets, and categorize them into three distinct groups: malignant, benign and ambiguous heterophilic datasets. We identify malignant and ambiguous heterophily as the truly challenging subsets of tasks, and to our best knowledge, we are the first to propose such taxonomy. Then, we re-evaluate $11$ state-of-the-arts (SOTA) GNNs, covering six popular methods, with fine-tuned hyperparameters on different groups of heterophilic datasets. Based on the model performance, we comprehensively reassess the effectiveness of different methods on heterophily. At last, we evaluate $11$ popular homophily metrics on synthetic graphs with three different graph generation approaches. To overcome the unreliability of observation-based comparison and evaluation, we conduct the first quantitative evaluation and provide detailed analysis.

Xinyu Wang, Lin Gui, Yulan He

Table of contents (ToC) extraction centres on structuring documents in a hierarchical manner. In this paper, we propose a new dataset, ESGDoc, comprising 1,093 ESG annual reports from 563 companies spanning from 2001 to 2022. These reports pose significant challenges due to their diverse structures and extensive length. To address these challenges, we propose a new framework for Toc extraction, consisting of three steps: (1) Constructing an initial tree of text blocks based on reading order and font sizes; (2) Modelling each tree node (or text block) independently by considering its contextual information captured in node-centric subtree; (3) Modifying the original tree by taking appropriate action on each tree node (Keep, Delete, or Move). This construction-modelling-modification (CMM) process offers several benefits. It eliminates the need for pairwise modelling of section headings as in previous approaches, making document segmentation practically feasible. By incorporating structured information, each section heading can leverage both local and long-distance context relevant to itself. Experimental results show that our approach outperforms the previous state-of-the-art baseline with a fraction of running time. Our framework proves its scalability by effectively handling documents of any length.

Zeyu Wang, Kudria Sergei, Jingbang Chen et al.

Signed graphs are widely used to analyze complex systems such as social, political, and biological networks. The notion of balance, a key concept of signed graphs, reflects the stability of relationships. While it has been extensively studied in deterministic graphs, real-world networks often exhibit uncertainty in their connections, which traditional approaches struggle to address. To bridge this gap, we introduce the concept of balance rate, a metric for quantifying the degree of balance in uncertain signed graphs, and prove that computing it exactly is NP-hard, motivating the need for efficient estimation methods. We propose a novel Rao-Blackwellized spanning-tree estimator that achieves near-linear time complexity per sample by leveraging graph decomposition and structural properties. We also construct asymptotically justified confidence intervals using the Delta method. Experiments on real-world datasets demonstrate the efficiency and effectiveness of our approach, enabling scalable balance analysis in uncertain signed graphs.

Xinyu Wang, Yohan Lee, Juneyoung Park

This paper surveys and organizes research works in an under-studied area, which we call automated evaluation for student argumentative writing. Unlike traditional automated writing evaluation that focuses on holistic essay scoring, this field is more specific: it focuses on evaluating argumentative essays and offers specific feedback, including argumentation structures, argument strength trait score, etc. The focused and detailed evaluation is useful for helping students acquire important argumentation skill. In this paper we organize existing works around tasks, data and methods. We further experiment with BERT on representative datasets, aiming to provide up-to-date baselines for this field.

Kuan Li, Zhongwang Zhang, Huifeng Yin et al.

Transcending human cognitive limitations represents a critical frontier in LLM training. Proprietary agentic systems like DeepResearch have demonstrated superhuman capabilities on extremely complex information-seeking benchmarks such as BrowseComp, a feat previously unattainable. We posit that their success hinges on a sophisticated reasoning pattern absent in open-source models: the ability to systematically reduce extreme uncertainty when navigating vast information landscapes. Based on this insight, we introduce WebSailor, a complete post-training methodology designed to instill this crucial capability. Our approach involves generating novel, high-uncertainty tasks through structured sampling and information obfuscation, RFT cold start, and an efficient agentic RL training algorithm, Duplicating Sampling Policy Optimization (DUPO). With this integrated pipeline, WebSailor significantly outperforms all open-source agents in complex information-seeking tasks, matching proprietary agents' performance and closing the capability gap.

Pengjie Wang, Kaile Zhang, Xinyu Wang et al.

Oracle bone script, one of the earliest known forms of ancient Chinese writing, presents invaluable research materials for scholars studying the humanities and geography of the Shang Dynasty, dating back 3,000 years. The immense historical and cultural significance of these writings cannot be overstated. However, the passage of time has obscured much of their meaning, presenting a significant challenge in deciphering these ancient texts. With the advent of Artificial Intelligence (AI), employing AI to assist in deciphering Oracle Bone Characters (OBCs) has become a feasible option. Yet, progress in this area has been hindered by a lack of high-quality datasets. To address this issue, this paper details the creation of the HUST-OBC dataset. This dataset encompasses 77,064 images of 1,588 individual deciphered characters and 62,989 images of 9,411 undeciphered characters, with a total of 140,053 images, compiled from diverse sources. The hope is that this dataset could inspire and assist future research in deciphering those unknown OBCs. All the codes and datasets are available at https://github.com/Yuliang-Liu/Open-Oracle.

Liangcai Su, Zhen Zhang, Guangyu Li et al.

Large language models (LLMs) have evolved into agentic systems capable of autonomous tool use and multi-step reasoning for complex problem-solving. However, post-training approaches building upon general-purpose foundation models consistently underperform in agentic tasks, particularly in open-source implementations. We identify the root cause: the absence of robust agentic foundation models forces models during post-training to simultaneously learn diverse agentic behaviors while aligning them to expert demonstrations, thereby creating fundamental optimization tensions. To this end, we are the first to propose incorporating Agentic Continual Pre-training (Agentic CPT) into the deep research agents training pipeline to build powerful agentic foundational models. Based on this approach, we develop a deep research agent model named AgentFounder. We evaluate our AgentFounder-30B on 10 benchmarks and achieve state-of-the-art performance while retains strong tool-use ability, notably 39.9% on BrowseComp-en, 43.3% on BrowseComp-zh, and 31.5% Pass@1 on HLE.

Xinyu Wang, Bohan Zhuang, Qi Wu

Large Vision Language Models (LVLMs) have demonstrated remarkable abilities in understanding and reasoning about both visual and textual information. However, existing evaluation methods for LVLMs, primarily based on benchmarks like Visual Question Answering and image captioning, often fail to capture the full scope of LVLMs' capabilities. These benchmarks are limited by issues such as inadequate assessment of detailed visual perception, data contamination, and a lack of focus on multi-turn reasoning. To address these challenges, we propose \method{}, a game-based evaluation framework designed to provide a comprehensive assessment of LVLMs' cognitive and reasoning skills in structured environments. \method{} uses a set of games to evaluate LVLMs on four core tasks: Perceiving, Question Answering, Rule Following, and End-to-End Playing, with each target task designed to assess specific abilities, including visual perception, reasoning, decision-making, etc. Based on this framework, we conduct extensive experiments that explore the limitations of current LVLMs, such as handling long structured outputs and perceiving detailed and dense elements. Code and data are publicly available at https://github.com/xinke-wang/LVLM-Playground.

Xiang Li, Xiangyu Zhou, Rui Dong et al.

We propose a new synthesis algorithm that can efficiently search programs with local variables (e.g., those introduced by lambdas). Prior bottom-up synthesis algorithms are not able to evaluate programs with free local variables, and therefore cannot effectively reduce the search space of such programs (e.g., using standard observational equivalence reduction techniques), making synthesis slow. Our algorithm can reduce the space of programs with local variables. The key idea, dubbed lifted interpretation, is to lift up the program interpretation process, from evaluating one program at a time to simultaneously evaluating all programs from a grammar. Lifted interpretation provides a mechanism to systematically enumerate all binding contexts for local variables, thereby enabling us to evaluate and reduce the space of programs with local variables. Our ideas are instantiated in the domain of web automation. The resulting tool, Arborist, can automate a significantly broader range of challenging tasks more efficiently than state-of-the-art techniques including WebRobot and Helena.

Zhen Zhang, Liangcai Su, Zhuo Chen et al.

Deep research agents have achieved remarkable progress on complex information seeking tasks. Even long ReAct style rollouts explore only a single trajectory, while recent state of the art systems scale inference time compute via parallel search and aggregation. Yet deep research answers are composed of complementary pieces of evidence, which parallel rollouts often duplicate rather than complete, yielding diminishing returns while pushing the aggregation context toward the model's limit. We propose Argus, an agentic system in which a Searcher and a Navigator cooperate to treat deep research as assembling a jigsaw from complementary evidence pieces, rather than brute forcing the whole answer in parallel. The Searcher collects evidence traces for a given sub-query through ReAct-style interaction. The Navigator maintains a shared evidence graph, verifying which pieces are still missing, dispatching Searchers to gather them, and reasoning over the completed graph to produce a source-traced final answer. We train the Navigator with reinforcement learning to verify, dispatch, and synthesize, while independently training the Searcher to remain a standard ReAct agent. The resulting Navigator supports rollouts with a single Searcher or many in parallel without retraining. With both Searcher and Navigator built on a 35B-A3B MoE backbone, Argus gains 5.5 points with a single Searcher and 12.7 points with 8 parallel Searchers, averaged over eight benchmarks. With 64 Searchers it reaches 86.2 on BrowseComp, surpassing every proprietary agent we benchmark, while the Navigator's reasoning context stays under 21.5K tokens.

Zhuo Chen, Xinyu Geng, Xinyu Wang et al.

Visual Question-Answering (VQA) is a challenging multimodal task that requires integrating visual and textual information to generate accurate responses. While multimodal Retrieval-Augmented Generation (mRAG) has shown promise in enhancing VQA systems by providing more evidence on both image and text sides, the default procedure that addresses VQA queries, especially the knowledge-intensive ones, often relies on multi-stage pipelines of mRAG with inherent dependencies. To mitigate the inefficiency limitations while maintaining VQA task performance, this paper proposes a method that trains a multimodal planning agent, dynamically decomposing the mRAG pipeline to solve the VQA task. Our method optimizes the trade-off between efficiency and effectiveness by training the agent to intelligently determine the necessity of each mRAG step. In our experiments, the agent can help reduce redundant computations, cutting search time by over 60\% compared to existing methods and decreasing costly tool calls. Meanwhile, experiments demonstrate that our method outperforms all baselines, including a Deep Research agent and a carefully designed prompt-based method, on average over six various datasets. Code will be released.

An Yang, Anfeng Li, Baosong Yang et al. · tsinghua

In this work, we present Qwen3, the latest version of the Qwen model family. Qwen3 comprises a series of large language models (LLMs) designed to advance performance, efficiency, and multilingual capabilities. The Qwen3 series includes models of both dense and Mixture-of-Expert (MoE) architectures, with parameter scales ranging from 0.6 to 235 billion. A key innovation in Qwen3 is the integration of thinking mode (for complex, multi-step reasoning) and non-thinking mode (for rapid, context-driven responses) into a unified framework. This eliminates the need to switch between different models--such as chat-optimized models (e.g., GPT-4o) and dedicated reasoning models (e.g., QwQ-32B)--and enables dynamic mode switching based on user queries or chat templates. Meanwhile, Qwen3 introduces a thinking budget mechanism, allowing users to allocate computational resources adaptively during inference, thereby balancing latency and performance based on task complexity. Moreover, by leveraging the knowledge from the flagship models, we significantly reduce the computational resources required to build smaller-scale models, while ensuring their highly competitive performance. Empirical evaluations demonstrate that Qwen3 achieves state-of-the-art results across diverse benchmarks, including tasks in code generation, mathematical reasoning, agent tasks, etc., competitive against larger MoE models and proprietary models. Compared to its predecessor Qwen2.5, Qwen3 expands multilingual support from 29 to 119 languages and dialects, enhancing global accessibility through improved cross-lingual understanding and generation capabilities. To facilitate reproducibility and community-driven research and development, all Qwen3 models are publicly accessible under Apache 2.0.