Xiaoyang Wang

Fuxiao Liu, Xiaoyang Wang, Wenlin Yao et al. · tencent-ai

With the rapid development of large language models (LLMs) and their integration into large multimodal models (LMMs), there has been impressive progress in zero-shot completion of user-oriented vision-language tasks. However, a gap remains in the domain of chart image understanding due to the distinct abstract components in charts. To address this, we introduce a large-scale MultiModal Chart Instruction (\textbf{MMC-Instruction}) dataset comprising 600k instances supporting diverse tasks and chart types. Leveraging this data, we develop MultiModal Chart Assistant (\textbf{MMCA}), an LMM that achieves state-of-the-art performance on existing chart QA benchmarks. Recognizing the need for a comprehensive evaluation of LMM chart understanding, we also propose a MultiModal Chart Benchmark (\textbf{MMC-Benchmark}), a comprehensive human-annotated benchmark with nine distinct tasks evaluating reasoning capabilities over charts. Extensive experiments on MMC-Benchmark reveal the limitations of existing LMMs on correctly interpreting charts, even for the most recent GPT-4V model. Our work provides an instruction-tuning methodology and benchmark to advance multimodal understanding of charts. Code and data are available at https://github.com/FuxiaoLiu/MMC.

Fei Wang, Kaiqiang Song, Hongming Zhang et al. · tencent-ai

Abstractive summarization models typically learn to capture the salient information from scratch implicitly. Recent literature adds extractive summaries as guidance for abstractive summarization models to provide hints of salient content and achieves better performance. However, extractive summaries as guidance could be over strict, leading to information loss or noisy signals. Furthermore, it cannot easily adapt to documents with various abstractiveness. As the number and allocation of salience content pieces vary, it is hard to find a fixed threshold deciding which content should be included in the guidance. In this paper, we propose a novel summarization approach with a flexible and reliable salience guidance, namely SEASON (SaliencE Allocation as Guidance for Abstractive SummarizatiON). SEASON utilizes the allocation of salience expectation to guide abstractive summarization and adapts well to articles in different abstractiveness. Automatic and human evaluations on two benchmark datasets show that the proposed method is effective and reliable. Empirical results on more than one million news articles demonstrate a natural fifteen-fifty salience split for news article sentences, providing a useful insight for composing news articles.

Sangwoo Cho, Kaiqiang Song, Xiaoyang Wang et al. · tencent-ai

Text segmentation is important for signaling a document's structure. Without segmenting a long document into topically coherent sections, it is difficult for readers to comprehend the text, let alone find important information. The problem is only exacerbated by a lack of segmentation in transcripts of audio/video recordings. In this paper, we explore the role that section segmentation plays in extractive summarization of written and spoken documents. Our approach learns robust sentence representations by performing summarization and segmentation simultaneously, which is further enhanced by an optimization-based regularizer to promote selection of diverse summary sentences. We conduct experiments on multiple datasets ranging from scientific articles to spoken transcripts to evaluate the model's performance. Our findings suggest that the model can not only achieve state-of-the-art performance on publicly available benchmarks, but demonstrate better cross-genre transferability when equipped with text segmentation. We perform a series of analyses to quantify the impact of section segmentation on summarizing written and spoken documents of substantial length and complexity.

Haopeng Zhang, Sangwoo Cho, Kaiqiang Song et al. · stanford, tencent-ai

Multi-document summarization aims to obtain core information from a collection of documents written on the same topic. This paper proposes a new holistic framework for unsupervised multi-document extractive summarization. Our method incorporates the holistic beam search inference method associated with the holistic measurements, named Subset Representative Index (SRI). SRI balances the importance and diversity of a subset of sentences from the source documents and can be calculated in unsupervised and adaptive manners. To demonstrate the effectiveness of our method, we conduct extensive experiments on both small and large-scale multi-document summarization datasets under both unsupervised and adaptive settings. The proposed method outperforms strong baselines by a significant margin, as indicated by the resulting ROUGE scores and diversity measures. Our findings also suggest that diversity is essential for improving multi-document summary performance.

Jiaao Chen, Xiaoman Pan, Dian Yu et al. · gatech, tencent-ai

We investigate how to elicit compositional generalization capabilities in large language models (LLMs). Compositional generalization empowers LLMs to solve complex problems by combining foundational skills, a critical reasoning ability akin to human intelligence. However, even the most advanced LLMs currently struggle with this form of reasoning. We examine this problem within the framework of in-context learning and find that demonstrating both foundational skills and compositional examples grounded in these skills within the same prompt context is crucial. We refer to this prompt structure as skills-in-context (SKiC). With as few as two exemplars, this in-context learning structure enables LLMs to tackle more challenging problems requiring innovative skill combinations, achieving near-perfect systematic generalization across a broad range of tasks. Intriguingly, SKiC also unlocks the latent potential of LLMs, allowing them to more actively utilize pre-existing internal skills acquired during earlier pretraining stages to solve complex reasoning problems. The SKiC structure is robust across different skill constructions and exemplar choices and demonstrates strong transferability to new tasks. Finally, inspired by our in-context learning study, we show that fine-tuning LLMs with SKiC-style data can elicit zero-shot weak-to-strong generalization, enabling the models to solve much harder problems directly with standard prompting.

Xianjun Yang, Kaiqiang Song, Sangwoo Cho et al. · tencent-ai

Aspect or query-based summarization has recently caught more attention, as it can generate differentiated summaries based on users' interests. However, the current dataset for aspect or query-based summarization either focuses on specific domains, contains relatively small-scale instances, or includes only a few aspect types. Such limitations hinder further explorations in this direction. In this work, we take advantage of crowd-sourcing knowledge on Wikipedia.org and automatically create a high-quality, large-scale open-domain aspect-based summarization dataset named OASum, which contains more than 3.7 million instances with around 1 million different aspects on 2 million Wikipedia pages. We provide benchmark results on OASum and demonstrate its ability for diverse aspect-based summarization generation. To overcome the data scarcity problem on specific domains, we also perform zero-shot, few-shot, and fine-tuning on seven downstream datasets. Specifically, zero/few-shot and fine-tuning results show that the model pre-trained on our corpus demonstrates a strong aspect or query-focused generation ability compared with the backbone model. Our dataset and pre-trained checkpoints are publicly available.

Yue Yang, Wenlin Yao, Hongming Zhang et al. · allen-ai

Large-scale pretrained language models have made significant advances in solving downstream language understanding tasks. However, they generally suffer from reporting bias, the phenomenon describing the lack of explicit commonsense knowledge in written text, e.g., ''an orange is orange''. To overcome this limitation, we develop a novel approach, Z-LaVI, to endow language models with visual imagination capabilities. Specifically, we leverage two complementary types of ''imaginations'': (i) recalling existing images through retrieval and (ii) synthesizing nonexistent images via text-to-image generation. Jointly exploiting the language inputs and the imagination, a pretrained vision-language model (e.g., CLIP) eventually composes a zero-shot solution to the original language tasks. Notably, fueling language models with imagination can effectively leverage visual knowledge to solve plain language tasks. In consequence, Z-LaVI consistently improves the zero-shot performance of existing language models across a diverse set of language tasks.

Xuansheng Wu, Wenlin Yao, Jianshu Chen et al.

Large Language Models (LLMs) have achieved remarkable success, where instruction tuning is the critical step in aligning LLMs with user intentions. In this work, we investigate how the instruction tuning adjusts pre-trained models with a focus on intrinsic changes. Specifically, we first develop several local and global explanation methods, including a gradient-based method for input-output attribution, and techniques for interpreting patterns and concepts in self-attention and feed-forward layers. The impact of instruction tuning is then studied by comparing the explanations derived from the pre-trained and instruction-tuned models. This approach provides an internal perspective of the model shifts on a human-comprehensible level. Our findings reveal three significant impacts of instruction tuning: 1) It empowers LLMs to recognize the instruction parts of user prompts, and promotes the response generation constantly conditioned on the instructions. 2) It encourages the self-attention heads to capture more word-word relationships about instruction verbs. 3) It encourages the feed-forward networks to rotate their pre-trained knowledge toward user-oriented tasks. These insights contribute to a more comprehensive understanding of instruction tuning and lay the groundwork for future work that aims at explaining and optimizing LLMs for various applications. Our code and data are publicly available at https://github.com/JacksonWuxs/Interpret_Instruction_Tuning_LLMs.

Kaiqiang Song, Chen Li, Xiaoyang Wang et al.

Podcasts have recently shown a rapid rise in popularity. Summarization of podcast transcripts is of practical benefit to both content providers and consumers. It helps consumers to quickly decide whether they will listen to the podcasts and reduces the cognitive load of content providers to write summaries. Nevertheless, podcast summarization faces significant challenges including factual inconsistencies with respect to the inputs. The problem is exacerbated by speech disfluencies and recognition errors in transcripts of spoken language. In this paper, we explore a novel abstractive summarization method to alleviate these challenges. Specifically, our approach learns to produce an abstractive summary while grounding summary segments in specific portions of the transcript to allow for full inspection of summary details. We conduct a series of analyses of the proposed approach on a large podcast dataset and show that the approach can achieve promising results. Grounded summaries bring clear benefits in locating the summary and transcript segments that contain inconsistent information, and hence significantly improve summarization quality in both automatic and human evaluation metrics.

Shanshan Han, Baturalp Buyukates, Zijian Hu et al.

This paper introduces FedSecurity, an end-to-end benchmark that serves as a supplementary component of the FedML library for simulating adversarial attacks and corresponding defense mechanisms in Federated Learning (FL). FedSecurity eliminates the need for implementing the fundamental FL procedures, e.g., FL training and data loading, from scratch, thus enables users to focus on developing their own attack and defense strategies. It contains two key components, including FedAttacker that conducts a variety of attacks during FL training, and FedDefender that implements defensive mechanisms to counteract these attacks. FedSecurity has the following features: i) It offers extensive customization options to accommodate a broad range of machine learning models (e.g., Logistic Regression, ResNet, and GAN) and FL optimizers (e.g., FedAVG, FedOPT, and FedNOVA); ii) it enables exploring the effectiveness of attacks and defenses across different datasets and models; and iii) it supports flexible configuration and customization through a configuration file and some APIs. We further demonstrate FedSecurity's utility and adaptability through federated training of Large Language Models (LLMs) to showcase its potential on a wide range of complex applications.

Robert J. Piechocki, Xiaoyang Wang, Mohammud J. Bocus

A new method for multimodal sensor fusion is introduced. The technique relies on a two-stage process. In the first stage, a multimodal generative model is constructed from unlabelled training data. In the second stage, the generative model serves as a reconstruction prior and the search manifold for the sensor fusion tasks. The method also handles cases where observations are accessed only via subsampling i.e. compressed sensing. We demonstrate the effectiveness and excellent performance on a range of multimodal fusion experiments such as multisensory classification, denoising, and recovery from subsampled observations.

Hui Wu, Xiaoyang Wang, Zhong Fan

Large language models (LLMs) are increasingly used to automate power-system analysis, but many utilities and energy-research labs require on-premise serving for confidentiality, regulatory, reproducibility, and cost reasons. This makes the reliability of open-weight models a deployment issue. We show that first-pass failures in power-system code generation are dominated not by reasoning alone, but by structured API-knowledge boundary errors: hallucinated function names, misused parameters, and mishandled result tables in versioned simulation libraries. We introduce PowerCodeBench, an execution-validated benchmark generator that pairs natural-language operator queries with pandapower code and numerical ground truth; an L0-L3 documentation-driven probing procedure that measures per-model API knowledge profiles; and a boundary-aware intervention that combines query-side API demand estimation with targeted proactive documentation injection and routed reactive correction. On a 2,000-task frozen release, we evaluate ten open-weight LLMs (1.5B-480B parameters) and four commercial mid-tier APIs. The intervention improves every evaluated open-weight model of at least 7B parameters and every commercial API by 32 to 56 accuracy points. Open-weight models in the 70B-120B range match the commercial mid-tier accuracy range, while Llama-3.1-405B and Qwen3-Coder-480B lead the panel. The targeted prompts preserve the full-context accuracy ceiling while using 41% of the prompt-token cost. The result is an accuracy-side, deployment-time path toward reliable on-premise LLM assistance for grid-analysis workflows without fine-tuning or cloud inference.

Miao Yan, Ping Zhang, Haofei Zhang et al.

Thermal infrared (TIR) tracking is pivotal in computer vision tasks due to its all-weather imaging capability. Traditional tracking methods predominantly rely on hand-crafted features, and while deep learning has introduced correlation filtering techniques, these are often constrained by rudimentary correlation operations. Furthermore, transformer-based approaches tend to overlook temporal and coordinate information, which is critical for TIR tracking that lacks texture and color information. In this paper, to address these issues, we apply natural language modeling to TIR tracking and propose a coordinate-aware thermal infrared tracking model called NLMTrack, which enhances the utilization of coordinate and temporal information. NLMTrack applies an encoder that unifies feature extraction and feature fusion, which simplifies the TIR tracking pipeline. To address the challenge of low detail and low contrast in TIR images, on the one hand, we design a multi-level progressive fusion module that enhances the semantic representation and incorporates multi-scale features. On the other hand, the decoder combines the TIR features and the coordinate sequence features using a causal transformer to generate the target sequence step by step. Moreover, we explore an adaptive loss aimed at elevating tracking accuracy and a simple template update strategy to accommodate the target's appearance variations. Experiments show that NLMTrack achieves state-of-the-art performance on multiple benchmarks. The Code is publicly available at \url{https://github.com/ELOESZHANG/NLMTrack}.

Zhen Zhang, Kaiqiang Song, Xun Wang et al.

AI agents are increasingly used to solve real-world tasks by reasoning over multi-turn user interactions and invoking external tools. However, applying reinforcement learning to such settings remains difficult: realistic objectives often lack verifiable rewards and instead emphasize open-ended behaviors; moreover, RL for multi-turn, multi-step agentic tool use is still underexplored; and building and maintaining executable tool environments is costly, limiting scale and coverage. We propose CM2, an RL framework that replaces verifiable outcome rewards with checklist rewards. CM2 decomposes each turn's intended behavior into fine-grained binary criteria with explicit evidence grounding and structured metadata, turning open-ended judging into more stable classification-style decisions. To balance stability and informativeness, our method adopts a strategy of sparse reward assignment but dense evaluation criteria. Training is performed in a scalable LLM-simulated tool environment, avoiding heavy engineering for large tool sets. Experiments show that CM2 consistently improves over supervised fine-tuning. Starting from an 8B Base model and training on an 8k-example RL dataset, CM2 improves over the SFT counterpart by 8 points on tau^-Bench, by 10 points on BFCL-V4, and by 12 points on ToolSandbox. The results match or even outperform similarly sized open-source baselines, including the judging model. CM2 thus provides a scalable recipe for optimizing multi-turn, multi-step tool-using agents without relying on verifiable rewards. Code provided by the open-source community: https://github.com/namezhenzhang/CM2-RLCR-Tool-Agent.

Hakan Erdol, Xiaoyang Wang, Peizheng Li et al.

The vision of 5G lies in providing high data rates, low latency (for the aim of near-real-time applications), significantly increased base station capacity, and near-perfect quality of service (QoS) for users, compared to LTE networks. In order to provide such services, 5G systems will support various combinations of access technologies such as LTE, NR, NR-U and Wi-Fi. Each radio access technology (RAT) provides different types of access, and these should be allocated and managed optimally among the users. Besides resource management, 5G systems will also support a dual connectivity service. The orchestration of the network therefore becomes a more difficult problem for system managers with respect to legacy access technologies. In this paper, we propose an algorithm for RAT allocation based on federated meta-learning (FML), which enables RAN intelligent controllers (RICs) to adapt more quickly to dynamically changing environments. We have designed a simulation environment which contains LTE and 5G NR service technologies. In the simulation, our objective is to fulfil UE demands within the deadline of transmission to provide higher QoS values. We compared our proposed algorithm with a single RL agent, the Reptile algorithm and a rule-based heuristic method. Simulation results show that the proposed FML method achieves higher caching rates at first deployment round 21% and 12% respectively. Moreover, proposed approach adapts to new tasks and environments most quickly amongst the compared methods.

Mengzhang Cai, Xin Gao, Yu Li et al.

The rapid evolution of Large Language Models (LLMs) is predicated on the quality and diversity of post-training datasets. However, a critical dichotomy persists: while models are rigorously benchmarked, the data fueling them remains a black box--characterized by opaque composition, uncertain provenance, and a lack of systematic evaluation. This opacity hinders reproducibility and obscures the causal link between data characteristics and model behaviors. To bridge this gap, we introduce OpenDataArena (ODA), a holistic and open platform designed to benchmark the intrinsic value of post-training data. ODA establishes a comprehensive ecosystem comprising four key pillars: (i) a unified training-evaluation pipeline that ensures fair, open comparisons across diverse models (e.g., Llama, Qwen) and domains; (ii) a multi-dimensional scoring framework that profiles data quality along tens of distinct axes; (iii) an interactive data lineage explorer to visualize dataset genealogy and dissect component sources; and (iv) a fully open-source toolkit for training, evaluation, and scoring to foster data research. Extensive experiments on ODA--covering over 120 training datasets across multiple domains on 22 benchmarks, validated by more than 600 training runs and 40 million processed data points--reveal non-trivial insights. Our analysis uncovers the inherent trade-offs between data complexity and task performance, identifies redundancy in popular benchmarks through lineage tracing, and maps the genealogical relationships across datasets. We release all results, tools, and configurations to democratize access to high-quality data evaluation. Rather than merely expanding a leaderboard, ODA envisions a shift from trial-and-error data curation to a principled science of Data-Centric AI, paving the way for rigorous studies on data mixing laws and the strategic composition of foundation models.

Liqiang Jing, Zhehui Huang, Xiaoyang Wang et al.

Large Language Models (LLMs) and Large Vision-Language Models (LVLMs) have demonstrated impressive language/vision reasoning abilities, igniting the recent trend of building agents for targeted applications such as shopping assistants or AI software engineers. Recently, many data science benchmarks have been proposed to investigate their performance in the data science domain. However, existing data science benchmarks still fall short when compared to real-world data science applications due to their simplified settings. To bridge this gap, we introduce DSBench, a comprehensive benchmark designed to evaluate data science agents with realistic tasks. This benchmark includes 466 data analysis tasks and 74 data modeling tasks, sourced from Eloquence and Kaggle competitions. DSBench offers a realistic setting by encompassing long contexts, multimodal task backgrounds, reasoning with large data files and multi-table structures, and performing end-to-end data modeling tasks. Our evaluation of state-of-the-art LLMs, LVLMs, and agents shows that they struggle with most tasks, with the best agent solving only 34.12% of data analysis tasks and achieving a 34.74% Relative Performance Gap (RPG). These findings underscore the need for further advancements in developing more practical, intelligent, and autonomous data science agents.

Zheng Liu, Honglin Lin, Chonghan Qin et al.

Chart reasoning is a critical capability for Vision Language Models (VLMs). However, the development of open-source models is severely hindered by the lack of high-quality training data. Existing datasets suffer from a dual challenge: synthetic charts are often simplistic and repetitive, while the associated QA pairs are prone to hallucinations and lack the reasoning depth required for complex tasks. To bridge this gap, we propose ChartVerse, a scalable framework designed to synthesize complex charts and reliable reasoning data from scratch. (1) To address the bottleneck of simple patterns, we first introduce Rollout Posterior Entropy (RPE), a novel metric that quantifies chart complexity. Guided by RPE, we develop complexity-aware chart coder to autonomously synthesize diverse, high-complexity charts via executable programs. (2) To guarantee reasoning rigor, we develop truth-anchored inverse QA synthesis. Diverging from standard generation, we adopt an answer-first paradigm: we extract deterministic answers directly from the source code, generate questions conditional on these anchors, and enforce strict consistency verification. To further elevate difficulty and reasoning depth, we filter samples based on model fail-rate and distill high-quality Chain-of-Thought (CoT) reasoning. We curate ChartVerse-SFT-600K and ChartVerse-RL-40K using Qwen3-VL-30B-A3B-Thinking as the teacher. Experimental results demonstrate that ChartVerse-8B achieves state-of-the-art performance, notably surpassing its teacher and rivaling the stronger Qwen3-VL-32B-Thinking.

Peizheng Li, Xiaoyang Wang, Robert Piechocki et al.

Measuring customer experience on mobile data is of utmost importance for global mobile operators. The reference signal received power (RSRP) is one of the important indicators for current mobile network management, evaluation and monitoring. Radio data gathered through the minimization of drive test (MDT), a 3GPP standard technique, is commonly used for radio network analysis. Collecting MDT data in different geographical areas is inefficient and constrained by the terrain conditions and user presence, hence is not an adequate technique for dynamic radio environments. In this paper, we study a generative model for RSRP prediction, exploiting MDT data and a digital twin (DT), and propose a data-driven, two-tier neural network (NN) model. In the first tier, environmental information related to user equipment (UE), base stations (BS) and network key performance indicators (KPI) are extracted through a variational autoencoder (VAE). The second tier is designed as a likelihood model. Here, the environmental features and real MDT data features are adopted, formulating an integrated training process. On validation, our proposed model that uses real-world data demonstrates an accuracy improvement of about 20% or more compared with the empirical model and about 10% when compared with a fully connected prediction network.

Mohammud J. Bocus, Xiaoyang Wang, Robert. J. Piechocki

This paper presents a novel approach for multimodal data fusion based on the Vector-Quantized Variational Autoencoder (VQVAE) architecture. The proposed method is simple yet effective in achieving excellent reconstruction performance on paired MNIST-SVHN data and WiFi spectrogram data. Additionally, the multimodal VQVAE model is extended to the 5G communication scenario, where an end-to-end Channel State Information (CSI) feedback system is implemented to compress data transmitted between the base-station (eNodeB) and User Equipment (UE), without significant loss of performance. The proposed model learns a discriminative compressed feature space for various types of input data (CSI, spectrograms, natural images, etc), making it a suitable solution for applications with limited computational resources.

Peizheng Li, Hakan Erdol, Keith Briggs et al.

Setting the transmit power setting of 5G cells has been a long-term topic of discussion, as optimized power settings can help reduce interference and improve the quality of service to users. Recently, machine learning (ML)-based, especially reinforcement learning (RL)-based control methods have received much attention. However, there is little discussion about the generalisation ability of the trained RL models. This paper points out that an RL agent trained in a specific indoor environment is room-dependent, and cannot directly serve new heterogeneous environments. Therefore, in the context of Open Radio Access Network (O-RAN), this paper proposes a distributed cell power-control scheme based on Federated Reinforcement Learning (FRL). Models in different indoor environments are aggregated to the global model during the training process, and then the central server broadcasts the updated model back to each client. The model will also be used as the base model for adaptive training in the new environment. The simulation results show that the FRL model has similar performance to a single RL agent, and both are better than the random power allocation method and exhaustive search method. The results of the generalisation test show that using the FRL model as the base model improves the convergence speed of the model in the new environment.

Peizheng Li, Jonathan Thomas, Xiaoyang Wang et al.

The next generation of networks will actively embrace artificial intelligence (AI) and machine learning (ML) technologies for automation networks and optimal network operation strategies. The emerging network structure represented by Open RAN (O-RAN) conforms to this trend, and the radio intelligent controller (RIC) at the centre of its specification serves as an ML applications host. Various ML models, especially Reinforcement Learning (RL) models, are regarded as the key to solving RAN-related multi-objective optimization problems. However, it should be recognized that most of the current RL successes are confined to abstract and simplified simulation environments, which may not directly translate to high performance in complex real environments. One of the main reasons is the modelling gap between the simulation and the real environment, which could make the RL agent trained by simulation ill-equipped for the real environment. This issue is termed as the sim2real gap. This article brings to the fore the sim2real challenge within the context of O-RAN. Specifically, it emphasizes the characteristics, and benefits that the digital twins (DT) could have as a place for model development and verification. Several use cases are presented to exemplify and demonstrate failure modes of the simulations trained RL model in real environments. The effectiveness of DT in assisting the development of RL algorithms is discussed. Then the current state of the art learning-based methods commonly used to overcome the sim2real challenge are presented. Finally, the development and deployment concerns for the RL applications realisation in O-RAN are discussed from the view of the potential issues like data interaction, environment bottlenecks, and algorithm design.

Xin Gao, Xiaoyang Wang, Yun Zhu et al.

The construction of Supervised Fine-Tuning (SFT) datasets is a critical yet under-theorized stage in the post-training of Large Language Models (LLMs), as prevalent practices often rely on heuristic aggregation without a systematic understanding of how individual samples contribute to model performance. In this report, we propose a paradigm shift from ad-hoc curation to a closed-loop dataset engineering framework using OpenDataArena (ODA), which leverages value-anchored rankings and multi-dimensional analysis to transform value benchmarking into feedback signals guiding dataset construction. We instantiate this methodology through two new datasets: \textbf{ODA-Math-460k}, a specialized mathematics reasoning dataset that utilizes a novel two-stage difficulty-aware pipeline to achieve State-of-the-Art (SOTA) results on benchmarks such as AIME and HMMT, and \textbf{ODA-Mixture (100k \& 500k)}, a series of multi-domain instruction datasets built via an ``Anchor-and-Patch'' strategy that outperforms significantly larger open-source baselines. Our empirical results demonstrate that ODA-driven datasets significantly improve both domain-specific reasoning and general utility while achieving superior data efficiency, validating a transition toward data-centric AI where transparent evaluation serves as the primary engine for engineering high-quality training data.

Xiaoyang Wang, Dimitrios Dimitriadis, Sanmi Koyejo et al.

Federated learning enables training high-utility models across several clients without directly sharing their private data. As a downside, the federated setting makes the model vulnerable to various adversarial attacks in the presence of malicious clients. Despite the theoretical and empirical success in defending against attacks that aim to degrade models' utility, defense against backdoor attacks that increase model accuracy on backdoor samples exclusively without hurting the utility on other samples remains challenging. To this end, we first analyze the failure modes of existing defenses over a flat loss landscape, which is common for well-designed neural networks such as Resnet (He et al., 2015) but is often overlooked by previous works. Then, we propose an invariant aggregator that redirects the aggregated update to invariant directions that are generally useful via selectively masking out the update elements that favor few and possibly malicious clients. Theoretical results suggest that our approach provably mitigates backdoor attacks and remains effective over flat loss landscapes. Empirical results on three datasets with different modalities and varying numbers of clients further demonstrate that our approach mitigates a broad class of backdoor attacks with a negligible cost on the model utility.

Qiuyu Zhu, Liang Zhang, Qianxiong Xu et al.

Despite the success of Heterogeneous Graph Neural Networks (HGNNs) in modeling real-world Heterogeneous Information Networks (HINs), challenges such as expressiveness limitations and over-smoothing have prompted researchers to explore Graph Transformers (GTs) for enhanced HIN representation learning. However, research on GT in HINs remains limited, with two key shortcomings in existing work: (1) A node's neighbors at different distances in HINs convey diverse semantics. Unfortunately, existing methods ignore such differences and uniformly treat neighbors within a given distance in a coarse manner, which results in semantic confusion. (2) Nodes in HINs have various types, each with unique semantics. Nevertheless, existing methods mix nodes of different types during neighbor aggregation, hindering the capture of proper correlations between nodes of diverse types. To bridge these gaps, we design an innovative structure named (k,t)-ring neighborhood, where nodes are initially organized by their distance, forming different non-overlapping k-ring neighborhoods for each distance. Within each k-ring structure, nodes are further categorized into different groups according to their types, thus emphasizing the heterogeneity of both distances and types in HINs naturally. Based on this structure, we propose a novel Hierarchical Heterogeneous Graph Transformer (HHGT) model, which seamlessly integrates a Type-level Transformer for aggregating nodes of different types within each k-ring neighborhood, followed by a Ring-level Transformer for aggregating different k-ring neighborhoods in a hierarchical manner. Extensive experiments are conducted on downstream tasks to verify HHGT's superiority over 14 baselines, with a notable improvement of up to 24.75% in NMI and 29.25% in ARI for node clustering task on the ACM dataset compared to the best baseline.

Xinqiao Zhao, Feilong Tang, Xiaoyang Wang et al.

Image-level weakly supervised semantic segmentation has received increasing attention due to its low annotation cost. Existing methods mainly rely on Class Activation Mapping (CAM) to obtain pseudo-labels for training semantic segmentation models. In this work, we are the first to demonstrate that long-tailed distribution in training data can cause the CAM calculated through classifier weights over-activated for head classes and under-activated for tail classes due to the shared features among head- and tail- classes. This degrades pseudo-label quality and further influences final semantic segmentation performance. To address this issue, we propose a Shared Feature Calibration (SFC) method for CAM generation. Specifically, we leverage the class prototypes that carry positive shared features and propose a Multi-Scaled Distribution-Weighted (MSDW) consistency loss for narrowing the gap between the CAMs generated through classifier weights and class prototypes during training. The MSDW loss counterbalances over-activation and under-activation by calibrating the shared features in head-/tail-class classifier weights. Experimental results show that our SFC significantly improves CAM boundaries and achieves new state-of-the-art performances. The project is available at https://github.com/Barrett-python/SFC.

Jianke Yu, Hanchen Wang, Chen Chen et al.

Graph Neural Networks (GNNs) are vital in data science but are increasingly susceptible to adversarial attacks. To help researchers develop more robust GNN models, it's essential to focus on designing strong attack models as foundational benchmarks and guiding references. Among adversarial attacks, gray-box poisoning attacks are noteworthy due to their effectiveness and fewer constraints. These attacks exploit GNNs' need for retraining on updated data, thereby impacting their performance by perturbing these datasets. However, current research overlooks the real-world scenario of incomplete graphs. To address this gap, we introduce the Robust Incomplete Deep Attack Framework (RIDA). It is the first algorithm for robust gray-box poisoning attacks on incomplete graphs. The approach innovatively aggregates distant vertex information and ensures powerful data utilization. Extensive tests against 9 SOTA baselines on 3 real-world datasets demonstrate that RIDA's superiority in handling incompleteness and high attack performance on the incomplete graph.

Xiaoyang Wang, Huihui Bai, Limin Yu et al.

Semi-supervised semantic segmentation allows model to mine effective supervision from unlabeled data to complement label-guided training. Recent research has primarily focused on consistency regularization techniques, exploring perturbation-invariant training at both the image and feature levels. In this work, we proposed a novel feature-level consistency learning framework named Density-Descending Feature Perturbation (DDFP). Inspired by the low-density separation assumption in semi-supervised learning, our key insight is that feature density can shed a light on the most promising direction for the segmentation classifier to explore, which is the regions with lower density. We propose to shift features with confident predictions towards lower-density regions by perturbation injection. The perturbed features are then supervised by the predictions on the original features, thereby compelling the classifier to explore less dense regions to effectively regularize the decision boundary. Central to our method is the estimation of feature density. To this end, we introduce a lightweight density estimator based on normalizing flow, allowing for efficient capture of the feature density distribution in an online manner. By extracting gradients from the density estimator, we can determine the direction towards less dense regions for each feature. The proposed DDFP outperforms other designs on feature-level perturbations and shows state of the art performances on both Pascal VOC and Cityscapes dataset under various partition protocols. The project is available at https://github.com/Gavinwxy/DDFP.

Yu Li, Xiaoran Shang, Qizhi Pei et al.

Post-training data plays a pivotal role in shaping the capabilities of Large Language Models (LLMs), yet datasets are often treated as isolated artifacts, overlooking the systemic connections that underlie their evolution. To disentangle these complex relationships, we introduce the concept of \textbf{data lineage} to the LLM ecosystem and propose an automated multi-agent framework to reconstruct the evolutionary graph of dataset development. Through large-scale lineage analysis, we characterize domain-specific structural patterns, such as vertical refinement in math-oriented datasets and horizontal aggregation in general-domain corpora. Moreover, we uncover pervasive systemic issues, including \textit{structural redundancy} induced by implicit dataset intersections and the \textit{propagation of benchmark contamination} along lineage paths. To demonstrate the practical value of lineage analysis for data construction, we leverage the reconstructed lineage graph to create a \textit{lineage-aware diversity-oriented dataset}. By anchoring instruction sampling at upstream root sources, this approach mitigates downstream homogenization and hidden redundancy, yielding a more diverse post-training corpus. We further highlight lineage-centric analysis as an efficient and robust topological alternative to sample-level dataset comparison for large-scale data ecosystems. By grounding data construction in explicit lineage structures, our work advances post-training data curation toward a more systematic and controllable paradigm.

Tong Zheng, Hongming Zhang, Wenhao Yu et al.

Parallel thinking has emerged as a novel approach for enhancing the reasoning capabilities of large language models (LLMs) by exploring multiple reasoning paths concurrently. However, activating such capabilities through training remains challenging, as existing methods predominantly rely on supervised fine-tuning (SFT) over synthetic data, which encourages teacher-forced imitation rather than exploration and generalization. Different from them, we propose \textbf{Parallel-R1}, the first reinforcement learning (RL) framework that enables parallel thinking behaviors for complex real-world reasoning tasks. Our framework employs a progressive curriculum that explicitly addresses the cold-start problem in training parallel thinking with RL. We first use SFT on prompt-generated trajectories from easier tasks to instill the parallel thinking ability, then transition to RL to explore and generalize this skill on harder problems. Experiments on various math benchmarks, including MATH, AMC23, and AIME, show that Parallel-R1 successfully instills parallel thinking, leading to 8.4% accuracy improvements over the sequential thinking model trained directly on challenging tasks with RL. Further analysis reveals a clear shift in the model's thinking behavior: at an early stage, it uses parallel thinking as an exploration strategy, while in a later stage, it uses the same capability for multi-perspective verification. Most significantly, we validate parallel thinking as a \textbf{mid-training exploration scaffold}, where this temporary exploratory phase unlocks a higher performance ceiling after RL, yielding a 42.9% improvement over the baseline on AIME25. Our model, data, and code will be open-source at https://github.com/zhengkid/Parallel-R1.

Yizheng Gong, Siyue Yu, Xiaoyang Wang et al.

Most continual segmentation methods tackle the problem as a per-pixel classification task. However, such a paradigm is very challenging, and we find query-based segmenters with built-in objectness have inherent advantages compared with per-pixel ones, as objectness has strong transfer ability and forgetting resistance. Based on these findings, we propose CoMasTRe by disentangling continual segmentation into two stages: forgetting-resistant continual objectness learning and well-researched continual classification. CoMasTRe uses a two-stage segmenter learning class-agnostic mask proposals at the first stage and leaving recognition to the second stage. During continual learning, a simple but effective distillation is adopted to strengthen objectness. To further mitigate the forgetting of old classes, we design a multi-label class distillation strategy suited for segmentation. We assess the effectiveness of CoMasTRe on PASCAL VOC and ADE20K. Extensive experiments show that our method outperforms per-pixel and query-based methods on both datasets. Code will be available at https://github.com/jordangong/CoMasTRe.

Tianqing Fang, Zhisong Zhang, Xiaoyang Wang et al. · tencent-ai

General AI Agents are increasingly recognized as foundational frameworks for the next generation of artificial intelligence, enabling complex reasoning, web interaction, coding, and autonomous research capabilities. However, current agent systems are either closed-source or heavily reliant on a variety of paid APIs and proprietary tools, limiting accessibility and reproducibility for the research community. In this work, we present \textbf{Cognitive Kernel-Pro}, a fully open-source and (to the maximum extent) free multi-module agent framework designed to democratize the development and evaluation of advanced AI agents. Within Cognitive Kernel-Pro, we systematically investigate the curation of high-quality training data for Agent Foundation Models, focusing on the construction of queries, trajectories, and verifiable answers across four key domains: web, file, code, and general reasoning. Furthermore, we explore novel strategies for agent test-time reflection and voting to enhance agent robustness and performance. We evaluate Cognitive Kernel-Pro on GAIA, achieving state-of-the-art results among open-source and free agents. Notably, our 8B-parameter open-source model surpasses previous leading systems such as WebDancer and WebSailor, establishing a new performance standard for accessible, high-capability AI agents. Code is available at https://github.com/Tencent/CognitiveKernel-Pro

Xiaoyang Wang, Xin Chen

The large-scale integration of inverter-interfaced renewable energy sources presents significant challenges to maintaining power balance and nominal frequency in modern power systems. This paper studies grid-level coordinated control of grid-forming (GFM) and grid-following (GFL) inverter-based resources (IBRs) for scalable and optimal frequency control. We propose a fully distributed optimal frequency control algorithm based on the projected primal-dual gradient method and by leveraging the structure of the underlying physical system dynamics. The proposed algorithm i) restores the nominal system frequency while minimizing total control cost and enforcing IBR power capacity limits and line thermal constraints, and ii) operates in a distributed manner that only needs local measurements and neighbor-to-neighbor communication. In particular, when the line thermal constraints are disregarded, the proposed algorithm admits a fully local implementation that requires no communication, while still ensuring optimality and satisfying IBR power capacity limits. We establish the global asymptotic convergence of the algorithm using Lyapunov stability analysis. The effectiveness and optimality of the proposed algorithms are validated through high-fidelity, 100% inverter-based electromagnetic transient (EMT) simulations on the IEEE 39-bus system.

Yixuan Li, Xuelin Liu, Xiaoyang Wang et al.

The ability to distinguish whether an image is generated by artificial intelligence (AI) is a crucial ingredient in human intelligence, usually accompanied by a complex and dialectical forensic and reasoning process. However, current fake image detection models and databases focus on binary classification without understandable explanations for the general populace. This weakens the credibility of authenticity judgment and may conceal potential model biases. Meanwhile, large multimodal models (LMMs) have exhibited immense visual-text capabilities on various tasks, bringing the potential for explainable fake image detection. Therefore, we pioneer the probe of LMMs for explainable fake image detection by presenting a multimodal database encompassing textual authenticity descriptions, the FakeBench. For construction, we first introduce a fine-grained taxonomy of generative visual forgery concerning human perception, based on which we collect forgery descriptions in human natural language with a human-in-the-loop strategy. FakeBench examines LMMs with four evaluation criteria: detection, reasoning, interpretation and fine-grained forgery analysis, to obtain deeper insights into image authenticity-relevant capabilities. Experiments on various LMMs confirm their merits and demerits in different aspects of fake image detection tasks. This research presents a paradigm shift towards transparency for the fake image detection area and reveals the need for greater emphasis on forensic elements in visual-language research and AI risk control. FakeBench will be available at https://github.com/Yixuan423/FakeBench.

Xiaoyang Wang, Christopher C. Yang

Large language models (LLMs) have revolutionized medical reasoning tasks, yet single-agent systems often falter on complex, interdisciplinary problems requiring robust handling of uncertainty and conflicting evidence. Multi-agent systems (MAS) leveraging LLMs enable collaborative intelligence, but prevailing centralized architectures suffer from scalability bottlenecks, single points of failure, and role confusion in resource-constrained environments. Decentralized MAS (D-MAS) promise enhanced autonomy and resilience via peer-to-peer interactions, but their application to high-stakes healthcare domains remains underexplored. We introduce MediHive, a novel decentralized multi-agent framework for medical question answering that integrates a shared memory pool with iterative fusion mechanisms. MediHive deploys LLM-based agents that autonomously self-assign specialized roles, conduct initial analyses, detect divergences through conditional evidence-based debates, and locally fuse peer insights over multiple rounds to achieve consensus. Empirically, MediHive outperforms single-LLM and centralized baselines on MedQA and PubMedQA datasets, attaining accuracies of 84.3% and 78.4%, respectively. Our work advances scalable, fault-tolerant D-MAS for medical AI, addressing key limitations of centralized designs while demonstrating superior performance in reasoning-intensive tasks.

Hakan Erdol, Xiaoyang Wang, Robert Piechocki et al.

The advancements of machine learning-based (ML) decision-making algorithms created various research and industrial opportunities. One of these areas is ML-based near-real-time network management applications (xApps) in Open-Radio Access Network (O-RAN). Normally, xApps are designed solely for the desired objectives, and fine-tuned for deployment. However, telecommunication companies can employ multiple xApps and deploy them in overlapping areas. Consider the different design objectives of xApps, the deployment might cause conflicts. To prevent such conflicts, we proposed the xApp distillation method that distills knowledge from multiple xApps, then uses this knowledge to train a single model that has retained the capabilities of Previous xApps. Performance evaluations show that compared conflict mitigation schemes can cause up to six times more network outages than xApp distillation in some cases.

Mengting Pan, Fan Li, Chen Chen et al.

Graph neural networks (GNNs) have achieved remarkable success in relational learning. However, their vulnerability to graph backdoor attacks (GBAs) poses a significant barrier to broader adoption in high-stakes applications. Despite recent advances in graph backdoor defense (GBD), existing methods primarily focus on subgraph-based GBAs, relying on the assumption that poisoned target nodes are explicitly connected to subgraph triggers. Our empirical results reveal that such structure-centric approaches fail to defend against emerging feature-based GBAs that preserve graph topology. Therefore, in this paper, we study a novel problem of universal graph backdoor defense. First, we investigate the shared effects of both attack types from a feature-based homophily perspective, which characterizes local feature consistency between nodes and their neighborhoods. Thorough theoretical and empirical analyses demonstrate that, regardless of trigger mechanisms, backdoors induced by GBAs exhibit lower feature-based homophily than clean nodes, indicating a discrepancy in local feature similarity. Motivated by this insight, we propose to leverage node-level local feature consistency, modeled by a neighbor-aware reconstruction loss, to distinguish backdoors from clean nodes. Then, a robust training strategy is developed to eliminate trigger effects while reducing noise induced by detection uncertainty. Extensive experiments demonstrate that our framework significantly degrades the attack success rate and maintains competitive clean accuracy under both subgraph-based and feature-based attacks.

Shwai He, Tao Ge, Guoheng Sun et al.

Traditional transformer models often allocate a fixed amount of computational resources to every input token, leading to inefficient and unnecessary computation. To address this, the Mixture of Depths (MoD) was introduced to dynamically adjust the computational depth by skipping less important layers. Despite its promise, current MoD approaches remain under-explored and face two main challenges: (1) high training costs due to the need to train the entire model along with the routers that determine which layers to skip, and (2) the risk of performance degradation when important layers are bypassed. In response to the first issue, we propose Router-Tuning, a method that fine-tunes only the router on a small dataset, drastically reducing the computational overhead associated with full model training. For the second challenge, we propose MindSkip, which deploys Attention with Dynamic Depths. This method preserves the model's performance while significantly enhancing computational and memory efficiency. Extensive experiments demonstrate that our approach delivers competitive results while dramatically improving the computation efficiency, e.g., 21\% speedup and only a 0.2\% performance drop. The code is released at https://github.com/CASE-Lab-UMD/Router-Tuning.

Xingyu Tan, Xiaoyang Wang, Qing Liu et al.

Retrieval-augmented generation (RAG) enhances large language models (LLMs) by incorporating external knowledge. Current hybrid RAG system retrieves evidence from both knowledge graphs (KGs) and text documents to support LLM reasoning. However, it faces challenges like handling multi-hop reasoning, multi-entity questions, multi-source verification, and effective graph utilization. To address these limitations, we present HydraRAG, a training-free framework that unifies graph topology, document semantics, and source reliability to support deep, faithful reasoning in LLMs. HydraRAG handles multi-hop and multi-entity problems through agent-driven exploration that combines structured and unstructured retrieval, increasing both diversity and precision of evidence. To tackle multi-source verification, HydraRAG uses a tri-factor cross-source verification (source trustworthiness assessment, cross-source corroboration, and entity-path alignment), to balance topic relevance with cross-modal agreement. By leveraging graph structure, HydraRAG fuses heterogeneous sources, guides efficient exploration, and prunes noise early. Comprehensive experiments on seven benchmark datasets show that HydraRAG achieves overall state-of-the-art results on all benchmarks with GPT-3.5-Turbo, outperforming the strong hybrid baseline ToG-2 by an average of 20.3% and up to 30.1%. Furthermore, HydraRAG enables smaller models (e.g., Llama-3.1-8B) to achieve reasoning performance comparable to that of GPT-4-Turbo. The source code is available on https://stevetantan.github.io/HydraRAG/.

Suncheng Xiang, Xiaoyang Wang, Junjie Jiang et al.

Colonoscopic Polyp Re-Identification aims to match the same polyp from a large gallery with images from different views taken using different cameras, which plays an important role in the prevention and treatment of colorectal cancer in computer-aided diagnosis. However, the coarse resolution of high-level features of a specific polyp often leads to inferior results for small objects where detailed information is important. To address this challenge, we propose a novel architecture, named Gated Progressive Fusion network, to selectively fuse features from multiple levels using gates in a fully connected way for polyp ReID. On the basis of it, a gated progressive fusion strategy is introduced to achieve layer-wise refinement of semantic information through multi-level feature interactions. Experiments on standard benchmarks show the benefits of the multimodal setting over state-of-the-art unimodal ReID models, especially when combined with the specialized multimodal fusion strategy.

Sijia Xu, Fan Li, Xiaoyang Wang et al.

Tabular data is the primary data format in industrial relational databases, underpinning modern data analytics and decision-making. However, the increasing scale of tabular data poses significant computational and storage challenges to learning-based analytical systems. This highlights the need for data-efficient learning, which enables effective model training and generalization using substantially fewer samples. Dataset condensation (DC) has emerged as a promising data-centric paradigm that synthesizes small yet informative datasets to preserve data utility while reducing storage and training costs. However, existing DC methods are computationally intensive due to reliance on complex gradient-based optimization. Moreover, they often overlook key characteristics of tabular data, such as heterogeneous features and class imbalance. To address these limitations, we introduce C$^{2}$TC (Class-Adaptive Clustering for Tabular Condensation), the first training-free tabular dataset condensation framework that jointly optimizes class allocation and feature representation, enabling efficient and scalable condensation. Specifically, we reformulate the dataset condensation objective into a novel class-adaptive cluster allocation problem (CCAP), which eliminates costly training and integrates adaptive label allocation to handle class imbalance. To solve the NP-hard CCAP, we develop HFILS, a heuristic local search that alternates between soft allocation and class-wise clustering to efficiently obtain high-quality solutions. Moreover, a hybrid categorical feature encoding (HCFE) is proposed for semantics-preserving clustering of heterogeneous discrete attributes. Extensive experiments on 10 real-world datasets demonstrate that C$^{2}$TC improves efficiency by at least 2 orders of magnitude over state-of-the-art baselines, while achieving superior downstream performance.

Qing Sima, Xiaoyang Wang, Wenjie Zhang

Community search aims to identify a refined set of nodes that are most relevant to a given query, supporting tasks ranging from fraud detection to recommendation. Unlike homophilic graphs, many real-world networks are heterophilic, where edges predominantly connect dissimilar nodes. Therefore, structural signals that once reflected smooth, low-frequency similarity now appear as sharp, high-frequency contrasts. However, both classical algorithms (e.g., k-core, k-truss) and recent ML-based models struggle to achieve effective community search on heterophilic graphs, where edge signs or semantics are generally unknown. Algorithm-based methods often return communities with mixed class labels, while GNNs, built on homophily, smooth away meaningful signals and blur community boundaries. Therefore, we propose Adaptive Community Search (AdaptCS), a lightweight framework featuring three key designs: (i) an AdaptCS Encoder that disentangles multi-hop and multi-frequency signals, enabling the model to capture both smooth (homophilic) and contrastive (heterophilic) relations; (ii) a memory-efficient low-rank optimization that removes the main computational bottleneck and ensures model scalability; and (iii) an Adaptive Community Score (ACS) that guides online search by balancing embedding similarity and topological relations. Extensive experiments on both heterophilic and homophilic benchmarks demonstrate that AdaptCS outperforms the best-performing baseline by an average of 11% in F1-score, retains robustness across heterophily levels, and achieves up to 2 orders of magnitude speedup over the strongest ML-based CS baselines.

Fan Li, Xiaoyang Wang, Chen Chen et al.

The increasing prevalence of large-scale hypergraphs poses significant computational challenges for hypergraph neural network (HNN) training. To address this, hypergraph condensation (HGC) distills large real hypergraphs into compact yet informative synthetic ones, beyond graph condensation (GC) methods limited to pairwise relations. However, existing HGC methods rely on decoupled training architectures, where structure generators are pre-trained on the original hypergraph but not jointly optimized with condensed features during refinement, resulting in misaligned structures that degrade downstream utility. Moreover, trajectory-based optimization incurs substantial computational overhead in refinement, limiting condensation efficiency. To tackle these issues, we propose \textbf{A}nchor-guided \textbf{H}yper\textbf{G}raph \textbf{C}ondensation with \textbf{D}ual-level \textbf{D}iscrimination (\textbf{AHGCDD}), which consists of three key components: (1) a node initialization module based on Heat Kernel PageRank (HKPR) to encode structural knowledge into feature semantics; (2) an anchor-guided hyperedge synthesis strategy for joint optimization of condensed features and structure; (3) a theoretically grounded dual-level discrimination objective for utility-preserving condensation without redundant HNN training. Extensive experiments demonstrate the superior effectiveness and efficiency of AHGCDD.

Xingyu Tan, Xiaoyang Wang, Qing Liu et al.

Knowledge graphs (KGs) provide structured evidence that can ground large language model (LLM) reasoning for knowledge-intensive question answering. However, many practical KGs are private, and sending retrieved triples or exploration traces to closed-source LLM APIs introduces leakage risk. Existing privacy treatments focus on masking entity names, but they still face four limitations: structural leakage under semantic masking, uncontrollable remote interaction, fragile multi-hop and multi-entity reasoning, and limited experience reuse for stability and efficiency. To address these issues, we propose PrivGemo, a privacy-preserving retrieval-augmented framework for KG-grounded reasoning with memory-guided exposure control. PrivGemo uses a dual-tower design to keep raw KG knowledge local while enabling remote reasoning over an anonymized view that goes beyond name masking to limit both semantic and structural exposure. PrivGemo supports multi-hop, multi-entity reasoning by retrieving anonymized long-hop paths that connect all topic entities, while keeping grounding and verification on the local KG. A hierarchical controller and a privacy-aware experience memory further reduce unnecessary exploration and remote interactions. Comprehensive experiments on six benchmarks show that PrivGemo achieves overall state-of-the-art results, outperforming the strongest baseline by up to 17.1%. Furthermore, PrivGemo enables smaller models (e.g., Qwen3-4B) to achieve reasoning performance comparable to that of GPT-4-Turbo.

Xinzhao Wang, Shuo Zhou, Xiaoyang Wang et al.

Trotter decomposition provides a simple approach to simulating open quantum systems by decomposing the Lindbladian into a sum of individual terms. While it is established that Trotter errors in Hamiltonian simulation depend on nested commutators of the summands, such a relationship remains poorly understood for Lindbladian dynamics. In this Letter, we derive commutator-based Trotter error bounds for Lindbladian simulation, yielding an $O(\sqrt{N})$ scaling in the number of Trotter steps for locally interacting systems on $N$ sites. When estimating observable averages, we apply Richardson extrapolation to achieve polylogarithmic precision while maintaining the commutator scaling. To bound the extrapolation remainder, we develop a general truncation bound for the Baker-Campbell-Hausdorff expansion that bypasses common convergence issues in physically relevant systems. For local Lindbladians, our results demonstrate that the Trotter-based methods outperform prior simulation techniques in system-size scaling while requiring only $O(1)$ ancillas. Numerical simulations further validate the predicted system-size and precision scaling.

Fan Li, Xiaoyang Wang, Wenjie Zhang et al.

Deep graph models have achieved great success in network representation learning. However, their focus on pairwise relationships restricts their ability to learn pervasive higher-order interactions in real-world systems, which can be naturally modeled as hypergraphs. To tackle this issue, Hypergraph Neural Networks (HNNs) have garnered substantial attention in recent years. Despite the proposal of numerous HNNs, the absence of consistent experimental protocols and multi-dimensional empirical analysis impedes deeper understanding and further development of HNN research. While several toolkits for deep hypergraph learning (DHGL) have been introduced to facilitate algorithm evaluation, they provide only limited quantitative evaluation results and insufficient coverage of advanced algorithms, datasets, and benchmark tasks. To fill the gap, we introduce DHG-Bench, the first comprehensive benchmark for HNNs. Specifically, DHG-Bench systematically investigates the characteristics of HNNs in terms of four dimensions: effectiveness, efficiency, robustness, and fairness. We comprehensively evaluate 17 state-of-the-art HNN algorithms on 22 diverse datasets spanning node-, edge-, and graph-level tasks, under unified experimental settings. Extensive experiments reveal both the strengths and limitations of existing algorithms, offering valuable insights and directions for future research. Furthermore, to facilitate reproducible research, we have developed an easy-to-use library for training and evaluating different HNN methods. The DHG-Bench library is available at: https://github.com/Coco-Hut/DHG-Bench.

Yebowen Hu, Kaiqiang Song, Sangwoo Cho et al.

Reasoning is most powerful when an LLM accurately aggregates relevant information. We examine the critical role of information aggregation in reasoning by requiring the LLM to analyze sports narratives. To succeed at this task, an LLM must infer points from actions, identify related entities, attribute points accurately to players and teams, and compile key statistics to draw conclusions. We conduct comprehensive experiments with real NBA basketball data and present SportsGen, a new method to synthesize game narratives. By synthesizing data, we can rigorously evaluate LLMs' reasoning capabilities under complex scenarios with varying narrative lengths and density of information. Our findings show that most models, including GPT-4o, often fail to accurately aggregate basketball scores due to frequent scoring patterns. Open-source models like Llama-3 further suffer from significant score hallucinations. Finally, the effectiveness of reasoning is influenced by narrative complexity, information density, and domain-specific terms, highlighting the challenges in analytical reasoning tasks.

Yiwei Qin, Kaiqiang Song, Yebowen Hu et al. · tencent-ai

This paper introduces the Decomposed Requirements Following Ratio (DRFR), a new metric for evaluating Large Language Models' (LLMs) ability to follow instructions. Addressing a gap in current methodologies, DRFR breaks down complex instructions into simpler criteria, facilitating a detailed analysis of LLMs' compliance with various aspects of tasks. Alongside this metric, we present InFoBench, a benchmark comprising 500 diverse instructions and 2,250 decomposed questions across multiple constraint categories. Our experiments compare DRFR with traditional scoring methods and explore annotation sources, including human experts, crowd-sourced workers, and GPT-4. The findings demonstrate DRFR's higher reliability and the effectiveness of using GPT-4 as a cost-efficient annotator. The evaluation of several advanced LLMs using this framework reveals their strengths and areas needing improvement, particularly in complex instruction-following. This study contributes a novel metric and benchmark, offering insights for future LLM development and evaluation.

Xiaoyang Wang, Yongkun Li, Lulu Yao et al.

RDMA link failures can render connections temporarily unavailable, causing both performance degradation and significant recovery overhead. To tolerate such failures, production datacenters assign each primary link with a standby link and, upon failure, uniformly retransmit all in-flight RDMA request over the backup path. However, we observe that such blanket retransmission is unnecessary. In-flight requests can be split into pre-failure and post-failure categories depending on whether the responder has already executed. Retransmitting post-failure requests is not only redundant (consuming bandwidth), but also incorrect for non-idempotent operations, where duplicate execution can violate application semantics. We present Varuna, a failure-type-aware RDMA recovery mechanism that enables correct retransmission and us-level failover. Varuna piggybacks a lightweight completion log on every RDMA operation; after a link failure, this log deterministically reveals which in-flight requests were executed (post-failure) and which were lost (pre-failure). Varuna then retransmits only the pre-failure subset and fetches/recovers the return values for post-failure requests. Evaluated using synthetic microbenchmarks and end-to-end RDMA TPC-C transactions, Varuna incurs only 0.6-10% steady-state latency overhead in realistic applications, eliminates 65% of recovery retransmission time, preserves transactional consistency, and introduces zero connectivity rebuild overhead and negligible memory overhead during RDMA failover.

Xingyu Tan, Xiaoyang Wang, Qing Liu et al.

Large Language Models (LLMs) have achieved impressive results in various tasks but struggle with hallucination problems and lack of relevant knowledge, especially in deep complex reasoning and knowledge-intensive tasks. Knowledge Graphs (KGs), which capture vast amounts of facts in a structured format, offer a reliable source of knowledge for reasoning. However, existing KG-based LLM reasoning methods face challenges like handling multi-hop reasoning, multi-entity questions, and effectively utilizing graph structures. To address these issues, we propose Paths-over-Graph (PoG), a novel method that enhances LLM reasoning by integrating knowledge reasoning paths from KGs, improving the interpretability and faithfulness of LLM outputs. PoG tackles multi-hop and multi-entity questions through a three-phase dynamic multi-hop path exploration, which combines the inherent knowledge of LLMs with factual knowledge from KGs. In order to improve the efficiency, PoG prunes irrelevant information from the graph exploration first and introduces efficient three-step pruning techniques that incorporate graph structures, LLM prompting, and a pre-trained language model (e.g., SBERT) to effectively narrow down the explored candidate paths. This ensures all reasoning paths contain highly relevant information captured from KGs, making the reasoning faithful and interpretable in problem-solving. PoG innovatively utilizes graph structure to prune the irrelevant noise and represents the first method to implement multi-entity deep path detection on KGs for LLM reasoning tasks. Comprehensive experiments on five benchmark KGQA datasets demonstrate PoG outperforms the state-of-the-art method ToG across GPT-3.5-Turbo and GPT-4, achieving an average accuracy improvement of 18.9%. Notably, PoG with GPT-3.5-Turbo surpasses ToG with GPT-4 by up to 23.9%.