arXiv subjects

Tong Xu

Publications and source records attributed to Tong Xu.

At least 55 records · Page 3Linked to original sources

arXiv2026

Mock Worlds, Real Skills: Building Small Agentic Language Models with Synthetic Tasks, Simulated Environments, and Rubric-Based Rewards

Small LLMs often struggle to match the agentic capabilities of large, costly models. While reinforcement learning can help, progress has been limited by two structural bottlenecks: existing open-source agentic training data are narrow in task variety and easily solved; real-world APIs lack diversity and are unstable for large-scale reinforcement learning rollout processes. We address these challenges with SYNTHAGENT, a framework that jointly synthesizes diverse tool-use training data and simulates complete environments. Specifically, a strong teacher model creates novel tasks and tool ecosystems, then rewrites them into intentionally underspecified instructions. This compels agents to actively query users for missing details. When handling synthetic tasks, an LLM-based user simulator provides user-private information, while a mock tool system delivers stable tool responses. For rewards, task-level rubrics are constructed based on required subgoals, user-agent interactions, and forbidden behaviors. Across 14 challenging datasets in math, search, and tool use, models trained on our synthetic data achieve substantial gains, with small models outperforming larger baselines.

cs.CL↗

arXiv2026

How Foundational Skills Influence VLM-based Embodied Agents:A Native Perspective

Recent advances in vision-language models (VLMs) have shown promise for human-level embodied intelligence. However, existing benchmarks for VLM-driven embodied agents often rely on high-level commands or discretized action spaces, which are non-native settings that differ markedly from real-world control. In addition, current benchmarks focus primarily on high-level tasks and lack joint evaluation and analysis at both low and high levels. To address these limitations, we present NativeEmbodied, a challenging benchmark for VLM-driven embodied agents that uses a unified, native low-level action space. Built on diverse simulated scenes, NativeEmbodied includes three representative high-level tasks in complex scenarios to evaluate overall performance. For more detailed analysis, we further decouple the skills required by complex tasks and construct four types of low-level tasks, each targeting a fundamental embodied skill. This joint evaluation across task and skill granularities enables fine-grained assessment of embodied agents. Experiments with state-of-the-art VLMs reveal clear deficiencies in several fundamental embodied skills, and further analysis shows that these bottlenecks significantly limit performance on high-level tasks. NativeEmbodied highlights key challenges for current VLM-driven embodied agents and provides insights to guide future research.

cs.AI↗

arXiv2026

PRIME: A Process-Outcome Alignment Benchmark for Verifiable Reasoning in Mathematics and Engineering

While model-based verifiers are essential for scaling Reinforcement Learning with Verifiable Rewards (RLVR), current outcome-centric verification paradigms primarily focus on the consistency between the final result and the ground truth, often neglecting potential errors in the derivation process. This leads to assigning positive rewards to correct answers produced from incorrect derivations. To bridge this gap, we introduce PRIME, a benchmark for evaluating verifiers on Process-Outcome Alignment verification in Mathematics and Engineering. Curated from a comprehensive collection of college-level STEM problems, PRIME comprises 2,530 high-difficulty samples through a consistency-based filtering pipeline. Through extensive evaluation, we find that current verifiers frequently fail to detect derivation flaws. Furthermore, we propose a process-aware RLVR training paradigm utilizing verifiers selected via PRIME. This approach substantially outperforms the outcome-only verification baseline, achieving absolute performance gains of 8.29%, 9.12%, and 7.31% on AIME24, AIME25, and Beyond-AIME, respectively, for the Qwen3-14B-Base model. Finally, we demonstrate a strong linear correlation ($R^2 > 0.92$) between verifier accuracy on PRIME and RLVR training effectiveness, validating PRIME as a reliable predictor for verifier selection.

cs.CL↗

arXiv2026

ChartAnchor: Chart Grounding with Structural-Semantic Fidelity

Recent advances in multimodal large language models (MLLMs) highlight the need for benchmarks that rigorously evaluate structured chart comprehension. Chart grounding refers to the bidirectional alignment between a chart's visual appearance and its structured semantics. This task requires models to produce a symbolic specification that faithfully captures the chart's visual and structural intent, while also recovering the underlying tabular data with precise values and relationships. Chart grounding directly reflects a model's capabilities in numerical reasoning, multimodal alignment, and structural reconstruction, and has several important real-world applications. Existing benchmarks, constrained by narrow chart diversity, isolated tasks, and incomplete evaluation frameworks, fail to holistically assess grounding. To address this, we propose ChartAnchor, a comprehensive benchmark of 8k+ chart-table-code triples spanning 30 chart types drawn from diverse real-world and augmented sources. ChartAnchor introduces two complementary tasks: chart-to-code generation and controlled chart-to-table reconstruction, enabling cross-validation of visual and numerical fidelity. A multi-level evaluation framework integrates semantic validation, stylistic analysis, and perceptual metrics to assess both structural and content-level correctness. Extensive experiments on MLLMs reveal critical limitations in numerical precision and code synthesis, emphasizing the need for structured reasoning beyond surface-level perception. By unifying symbolic and data-driven grounding, ChartAnchor establishes a rigorous foundation for chart grounding, offering meaningful insights for advancing MLLMs in scientific, financial, and industrial domains.

cs.AI↗

arXiv2026

Token-level Collaborative Alignment for LLM-based Generative Recommendation

Large Language Models (LLMs) have demonstrated strong potential for generative recommendation by leveraging rich semantic knowledge. However, existing LLM-based recommender systems struggle to effectively incorporate collaborative filtering (CF) signals, due to a fundamental mismatch between item-level preference modeling in CF and token-level next-token prediction (NTP) optimization in LLMs. Prior approaches typically treat CF as contextual hints or representation bias, and resort to multi-stage training to reduce behavioral semantic space discrepancies, leaving CF unable to explicitly regulate LLM generation. In this work, we propose Token-level Collaborative Alignment for Recommendation (TCA4Rec), a model-agnostic and plug-and-play framework that establishes an explicit optimization-level interface between CF supervision and LLM generation. TCA4Rec consists of (i) Collaborative Tokenizer, which projects raw item-level CF logits into token-level distributions aligned with the LLM token space, and (ii) Soft Label Alignment, which integrates these CF-informed distributions with one-hot supervision to optimize a soft NTP objective. This design preserves the generative nature of LLM training while enabling collaborative alignment with essential user preference of CF models. We highlight TCA4Rec is compatible with arbitrary traditional CF models and generalizes across a wide range of decoder-based LLM recommender architectures. Moreover, it provides an explicit mechanism to balance behavioral alignment and semantic fluency, yielding generative recommendations that are both accurate and controllable. Extensive experiments demonstrate that TCA4Rec consistently improves recommendation performance across a broad spectrum of CF models and LLM-based recommender systems.

cs.IR↗

arXiv2026

From Tags to Trees: Structuring Fine-Grained Knowledge for Controllable Data Selection in LLM Instruction Tuning

Effective and controllable data selection is critical for LLM instruction tuning, especially with massive open-source datasets. Existing approaches primarily rely on instance-level quality scores, or diversity metrics based on embedding clusters or semantic tags. However, constrained by the flatness of embedding spaces or the coarseness of tags, these approaches overlook fine-grained knowledge and its intrinsic hierarchical dependencies, consequently hindering precise data valuation and knowledge-aligned sampling. To address this challenge, we propose Tree-aware Aligned Global Sampling (TAGS), a unified framework that leverages a knowledge tree built from fine-grained tags, thereby enabling joint control of global quality, diversity, and target alignment. Using an LLM-based tagger, we extract atomic knowledge concepts, which are organized into a global tree through bottom-up hierarchical clustering. By grounding data instances onto this tree, a tree-aware metric then quantifies data quality and diversity, facilitating effective sampling. Our controllable sampling strategy maximizes tree-level information gain and enforces leaf-level alignment via KL-divergence for specific domains. Extensive experiments demonstrate that TAGS significantly outperforms state-of-the-art baselines. Notably, it surpasses the full-dataset model by \textbf{+5.84\%} using only \textbf{5\%} of the data, while our aligned sampling strategy further boosts average performance by \textbf{+4.24\%}.

cs.CL↗

arXiv2026

VIGIL: Defending LLM Agents Against Tool Stream Injection via Verify-Before-Commit

LLM agents operating in open environments face escalating risks from indirect prompt injection, particularly within the tool stream where manipulated metadata and runtime feedback hijack execution flow. Existing defenses encounter a critical dilemma as advanced models prioritize injected rules due to strict alignment while static protection mechanisms sever the feedback loop required for adaptive reasoning. To reconcile this conflict, we propose \textbf{VIGIL}, a framework that shifts the paradigm from restrictive isolation to a verify-before-commit protocol. By facilitating speculative hypothesis generation and enforcing safety through intent-grounded verification, \textbf{VIGIL} preserves reasoning flexibility while ensuring robust control. We further introduce \textbf{SIREN}, a benchmark comprising 959 tool stream injection cases designed to simulate pervasive threats characterized by dynamic dependencies. Extensive experiments demonstrate that \textbf{VIGIL} outperforms state-of-the-art dynamic defenses by reducing the attack success rate by over 22\% while more than doubling the utility under attack compared to static baselines, thereby achieving an optimal balance between security and utility.

cs.CR↗

arXiv2026

Align-GRAG: Anchor and Rationale Guided Dual Alignment for Graph Retrieval-Augmented Generation

Despite the strong abilities, large language models (LLMs) still suffer from hallucinations and reliance on outdated knowledge, raising concerns in knowledge-intensive tasks. Graph-based retrieval-augmented generation (GRAG) enriches LLMs with knowledge by retrieving graphs leveraging relational evidence, but it faces two challenges: structure-coupled irrelevant knowledge introduced by neighbor expansion and structure-reasoning discrepancy between graph embeddings and LLM semantics. We propose \ourmodel, an anchor-and-rationale guided refinement framework to address these challenges. It prompts an LLM to extract anchors and rationale chains, which provide intermediate supervision for \textbf{(1) node-level alignment} that identifies critical nodes and prunes noisy evidence, and \textbf{(2) graph-level alignment} that bridges graph and language semantic spaces via contrastive learning. Extensive experiments on commonsense reasoning, scene graph understanding, and knowledge graph reasoning demonstrate consistent gains over 18 strong baselines, validating the effectiveness of \ourmodel for improving graph-grounded generation. The code can be found in https://anonymous.4open.science/r/Align-GRAG-F3D8/.

cs.CL↗

arXiv2025

DMP-TTS: Disentangled multi-modal Prompting for Controllable Text-to-Speech with Chained Guidance

Controllable text-to-speech (TTS) systems face significant challenges in achieving independent manipulation of speaker timbre and speaking style, often suffering from entanglement between these attributes. We present DMP-TTS, a latent Diffusion Transformer (DiT) framework with explicit disentanglement and multi-modal prompting. A CLAP-based style encoder (Style-CLAP) aligns cues from reference audio and descriptive text in a shared space and is trained with contrastive learning plus multi-task supervision on style attributes. For fine-grained control during inference, we introduce chained classifier-free guidance (cCFG) trained with hierarchical condition dropout, enabling independent adjustment of content, timbre, and style guidance strengths. Additionally, we employ Representation Alignment (REPA) to distill acoustic-semantic features from a pretrained Whisper model into intermediate DiT representations, stabilizing training and accelerating convergence. Experiments show that DMP-TTS delivers stronger style controllability than open-source baselines while maintaining competitive intelligibility and naturalness. Code and demos will be available at https://y61329697.github.io/DMP-TTS/.

cs.SD↗

arXiv2025

MELLM: A Flow-Guided Large Language Model for Micro-Expression Understanding

Micro-expressions (MEs), brief and low-intensity facial movements revealing concealed emotions, are crucial for affective computing. Despite notable progress in ME recognition, existing methods are largely confined to discrete emotion classification, lacking the capacity for comprehensive ME Understanding (MEU), particularly in interpreting subtle facial dynamics and underlying emotional cues. While Multimodal Large Language Models (MLLMs) offer potential for MEU with their advanced reasoning abilities, they still struggle to perceive such subtle facial affective behaviors. To bridge this gap, we propose a ME Large Language Model (MELLM) that integrates optical flow-based sensitivity to subtle facial motions with the powerful inference ability of LLMs. Specifically, an iterative, warping-based optical-flow estimator, named MEFlowNet, is introduced to precisely capture facial micro-movements. For its training and evaluation, we construct MEFlowDataset, a large-scale optical-flow dataset with 54,611 onset-apex image pairs spanning diverse identities and subtle facial motions. Subsequently, we design a Flow-Guided Micro-Expression Understanding paradigm. Under this framework, the optical flow signals extracted by MEFlowNet are leveraged to build MEU-Instruct, an instruction-tuning dataset for MEU. MELLM is then fine-tuned on MEU-Instruct, enabling it to translate subtle motion patterns into human-readable descriptions and generate corresponding emotional inferences. Experiments demonstrate that MEFlowNet significantly outperforms existing optical flow methods in facial and ME-flow estimation, while MELLM achieves state-of-the-art accuracy and generalization across multiple ME benchmarks. To the best of our knowledge, this work presents two key contributions: MEFlowNet as the first dedicated ME flow estimator, and MELLM as the first LLM tailored for MEU.

cs.CV↗

arXiv2025

MemOS: A Memory OS for AI System

Large Language Models (LLMs) have become an essential infrastructure for Artificial General Intelligence (AGI), yet their lack of well-defined memory management systems hinders the development of long-context reasoning, continual personalization, and knowledge consistency.Existing models mainly rely on static parameters and short-lived contextual states, limiting their ability to track user preferences or update knowledge over extended periods.While Retrieval-Augmented Generation (RAG) introduces external knowledge in plain text, it remains a stateless workaround without lifecycle control or integration with persistent representations.Recent work has modeled the training and inference cost of LLMs from a memory hierarchy perspective, showing that introducing an explicit memory layer between parameter memory and external retrieval can substantially reduce these costs by externalizing specific knowledge. Beyond computational efficiency, LLMs face broader challenges arising from how information is distributed over time and context, requiring systems capable of managing heterogeneous knowledge spanning different temporal scales and sources. To address this challenge, we propose MemOS, a memory operating system that treats memory as a manageable system resource. It unifies the representation, scheduling, and evolution of plaintext, activation-based, and parameter-level memories, enabling cost-efficient storage and retrieval. As the basic unit, a MemCube encapsulates both memory content and metadata such as provenance and versioning. MemCubes can be composed, migrated, and fused over time, enabling flexible transitions between memory types and bridging retrieval with parameter-based learning. MemOS establishes a memory-centric system framework that brings controllability, plasticity, and evolvability to LLMs, laying the foundation for continual learning and personalized modeling.

cs.CL↗

arXiv2025

Look as You Think: Unifying Reasoning and Visual Evidence Attribution for Verifiable Document RAG via Reinforcement Learning

Aiming to identify precise evidence sources from visual documents, visual evidence attribution for visual document retrieval-augmented generation (VD-RAG) ensures reliable and verifiable predictions from vision-language models (VLMs) in multimodal question answering. Most existing methods adopt end-to-end training to facilitate intuitive answer verification. However, they lack fine-grained supervision and progressive traceability throughout the reasoning process. In this paper, we introduce the Chain-of-Evidence (CoE) paradigm for VD-RAG. CoE unifies Chain-of-Thought (CoT) reasoning and visual evidence attribution by grounding reference elements in reasoning steps to specific regions with bounding boxes and page indexes. To enable VLMs to generate such evidence-grounded reasoning, we propose Look As You Think (LAT), a reinforcement learning framework that trains models to produce verifiable reasoning paths with consistent attribution. During training, LAT evaluates the attribution consistency of each evidence region and provides rewards only when the CoE trajectory yields correct answers, encouraging process-level self-verification. Experiments on vanilla Qwen2.5-VL-7B-Instruct with Paper- and Wiki-VISA benchmarks show that LAT consistently improves the vanilla model in both single- and multi-image settings, yielding average gains of 8.23% in soft exact match (EM) and 47.0% in IoU@0.5. Meanwhile, LAT not only outperforms the supervised fine-tuning baseline, which is trained to directly produce answers with attribution, but also exhibits stronger generalization across domains.

cs.AI↗

arXiv2025

Xiangqi-R1: Enhancing Spatial Strategic Reasoning in LLMs for Chinese Chess via Reinforcement Learning

Game playing has long served as a fundamental benchmark for evaluating Artificial General Intelligence. While Large Language Models (LLMs) have demonstrated impressive capabilities in general reasoning, their effectiveness in spatial strategic reasoning, which is critical for complex and fully observable board games, remains insufficiently explored. In this work, we adopt Chinese Chess (Xiangqi) as a challenging and rich testbed due to its intricate rules and spatial complexity. To advance LLMs' strategic competence in such environments, we propose a training framework tailored to Xiangqi, built upon a large-scale dataset of five million board-move pairs enhanced with expert annotations and engine evaluations. Building on this foundation, we introduce Xiangqi-R1, a 7B-parameter model trained in multi-stage manner. Our Experimental results indicate that, despite their size and power, general-purpose LLMs struggle to achieve satisfactory performance in these tasks. Compared to general-purpose LLMs, Xiangqi-R1 greatly advances with an 18% rise in move legality and a 22% boost in analysis accuracy. Our results point to a promising path for creating general strategic intelligence in complex areas.

cs.AI↗

arXiv2025

Verti-Arena: A Controllable and Standardized Indoor Testbed for Multi-Terrain Off-Road Autonomy

Off-road navigation is an important capability for mobile robots deployed in environments that are inaccessible or dangerous to humans, such as disaster response or planetary exploration. Progress is limited due to the lack of a controllable and standardized real-world testbed for systematic data collection and validation. To fill this gap, we introduce Verti-Arena, a reconfigurable indoor facility designed specifically for off-road autonomy. By providing a repeatable benchmark environment, Verti-Arena supports reproducible experiments across a variety of vertically challenging terrains and provides precise ground truth measurements through onboard sensors and a motion capture system. Verti-Arena also supports consistent data collection and comparative evaluation of algorithms in off-road autonomy research. We also develop a web-based interface that enables research groups worldwide to remotely conduct standardized off-road autonomy experiments on Verti-Arena.

cs.RO↗

arXiv2025

Continuous Dynamic Modeling via Neural ODEs for Popularity Trajectory Prediction

Popularity prediction for information cascades has significant applications across various domains, including opinion monitoring and advertising recommendations. While most existing methods consider this as a discrete problem, popularity actually evolves continuously, exhibiting rich dynamic properties such as change rates and growth patterns. In this paper, we argue that popularity trajectory prediction is more practical, as it aims to forecast the entire trajectory of how popularity unfolds over arbitrary future time. This approach offers insights into both instantaneous popularity and the underlying dynamic properties. However, traditional methods for popularity trajectory prediction primarily rely on specific diffusion mechanism assumptions, which may not align well with real-world dynamics and compromise their performance. To address these limitations, we propose NODEPT, a novel approach based on neural ordinary differential equations (ODEs) for popularity trajectory prediction. NODEPT models the continuous dynamics of the underlying diffusion system using neural ODEs. We first employ an encoder to initialize the latent state representations of information cascades, consisting of two representation learning modules that capture the co-evolution structural characteristics and temporal patterns of cascades from different perspectives. More importantly, we then introduce an ODE-based generative module that learns the dynamics of the diffusion system in the latent space. Finally, a decoder transforms the latent state into the prediction of the future popularity trajectory. Our experimental results on three real-world datasets demonstrate the superiority and rationality of the proposed NODEPT method.

cs.SI↗

arXiv2025

A Survey on Parallel Reasoning

With the increasing capabilities of Large Language Models (LLMs), parallel reasoning has emerged as a new inference paradigm that enhances reasoning robustness by concurrently exploring multiple lines of thought before converging on a final answer. It has become a significant trend to explore parallel reasoning to overcome the fragility of standard sequential methods and improve practical performance. In this paper, we aim to survey and summarize the progress and challenges of parallel reasoning. We first present a formal definition of parallel reasoning and clarify its distinction from related concepts like Chain-of-Thought. Then, we organize and discuss advanced techniques based on a novel taxonomy, including non-interactive reasoning, interactive reasoning, and efficiency-focused decoding strategies. Additionally, we explore various application scenarios, such as solving complex problems and enhancing the reliability of LLM outputs.Finally, we highlight the core challenges of parallel reasoning and suggest potential directions for future research. We hope that our work can provide a useful roadmap for beginners and encourage more research on improving parallel reasoning methods. Related source can be avaliable in https://github.com/PPPP-kaqiu/Awesome-Parallel-Reasoning.

cs.CL↗

arXiv2025

An Asymptotically Optimal Coordinate Descent Algorithm for Learning Bayesian Networks from Gaussian Models

This paper studies the problem of learning Bayesian networks from continuous observational data, generated according to a linear Gaussian structural equation model. We consider an $\ell_0$-penalized maximum likelihood estimator for this problem which is known to have favorable statistical properties but is computationally challenging to solve, especially for medium-sized Bayesian networks. We propose a new coordinate descent algorithm to approximate this estimator and prove several remarkable properties of our procedure: the algorithm converges to a coordinate-wise minimum, and despite the non-convexity of the loss function, as the sample size tends to infinity, the objective value of the coordinate descent solution converges to the optimal objective value of the $\ell_0$-penalized maximum likelihood estimator. Finite-sample statistical consistency guarantees are also established. To the best of our knowledge, our proposal is the first coordinate descent procedure endowed with optimality and statistical guarantees in the context of learning Bayesian networks. Numerical experiments on synthetic and real data demonstrate that our coordinate descent method can obtain near-optimal solutions while being scalable.

stat.ML↗

arXiv2025

Adaptive Dynamics Planning for Robot Navigation

Autonomous robot navigation systems often rely on hierarchical planning, where global planners compute collision-free paths without considering dynamics, and local planners enforce dynamics constraints to produce executable commands. This discontinuity in dynamics often leads to trajectory tracking failure in highly constrained environments. Recent approaches integrate dynamics within the entire planning process by gradually decreasing its fidelity, e.g., increasing integration steps and reducing collision checking resolution, for real-time planning efficiency. However, they assume that the fidelity of the dynamics should decrease according to a manually designed scheme. Such static settings fail to adapt to environmental complexity variations, resulting in computational overhead in simple environments or insufficient dynamics consideration in obstacle-rich scenarios. To overcome this limitation, we propose Adaptive Dynamics Planning (ADP), a learning-augmented paradigm that uses reinforcement learning to dynamically adjust robot dynamics properties, enabling planners to adapt across diverse environments. We integrate ADP into three different planners and further design a standalone ADP-based navigation system, benchmarking them against other baselines. Experiments in both simulation and real-world tests show that ADP consistently improves navigation success, safety, and efficiency.

cs.RO↗