arXiv subjects

Minrui Xu

Publications and source records attributed to Minrui Xu.

At least 19 recordsLinked to original sources

arXiv2026

Prune-OPD: Efficient and Reliable On-Policy Distillation for Long-Horizon Reasoning

On-policy distillation (OPD) leverages dense teacher rewards to enhance reasoning models. However, scaling OPD to long-horizon tasks exposes a critical flaw: as the student's generated prefix inevitably diverges from the teacher's thought process, the teacher's dense reward loses local exploitability. Continuing to generate and evaluate tokens on these ``drifted'' trajectories not only degrades reward quality but also incurs massive computational waste. To address this, we introduce \textbf{Prune-OPD}, a framework that dynamically aligns training budgets with supervision quality. By continuously monitoring the local compatibility between student and teacher predictions (e.g., via top-$k$ overlap), Prune-OPD detects prefix-drift events in real time. Upon detecting severe drift, it monotonically down-weights subsequent unreliable rewards and triggers dynamic rollout truncation. This allows the training process to halt futile generation and reallocate compute strictly to reliable teacher supervision. Across diverse teacher-student combinations, Prune-OPD consistently aligns computation with supervision reliability. When prefix drift makes dense teacher rewards unreliable, it reduces training time by 37.6\%--68.0\% while preserving, and often improving, performance on challenging benchmarks (AMC, AIME, HMMT). When student-teacher compatibility remains high, it automatically preserves long-context supervision by expanding the training window. These results suggest that Prune-OPD improves OPD not by blindly shortening rollouts, but by reallocating computation toward locally exploitable teacher rewards.

cs.LG↗

arXiv2026

EnvFactory: Scaling Tool-Use Agents via Executable Environments Synthesis and Robust RL

Equipping LLMs with tool-use capabilities via Agentic Reinforcement Learning (Agentic RL) is bottlenecked by two challenges: the lack of scalable, robust execution environments and the scarcity of realistic training data that captures implicit human reasoning. Existing approaches depend on costly real-world APIs, hallucination-prone LLM simulators, or synthetic environments that are often single-turn or depend on pre-collected documents. Moreover, synthetic trajectories are frequently over-specified, resembling instruction sequences rather than natural human intents, reducing their effectiveness for RL training. We introduce EnvFactory, a fully automated framework that addresses both challenges. EnvFactory autonomously explores and verifies stateful, executable tool environments from authentic resources, and synthesizes natural multi-turn trajectories through topology-aware sampling and calibrated refinement, producing grounded queries with implicit intents. Using only 85 verified environments across 7 domains, EnvFactory generates 2,575 SFT and RL trajectories. Despite using significantly fewer environments than prior work, which are often 5 times more, EnvFactory achieves superior training efficiency and downstream performance, improving Qwen3-series models by up to +15% on BFCLv3, +8.6% on MCP-Atlas, and +6% on conversational benchmarks including $τ^2$-Bench and VitaBench. By fully automating both environment construction and trajectory synthesis, EnvFactory provides a scalable, extensible, and robust foundation for Agentic RL.

cs.CL↗

arXiv2026

Split and Aggregation Learning for Foundation Models Over Mobile Embodied AI Network (MEAN): A Comprehensive Survey

The rapid advancements in foundation models and sixth-generation (6G) wireless communication systems necessitate the development of efficient, scalable, and privacy-preserving machine learning approaches. For foundation models in 6G, split learning (SL) and aggregation learning (AL) have emerged as promising paradigms that address key challenges in distributed artificial intelligence (AI), such as communication efficiency, resource allocation, and data privacy. SL enables multiple entities to collaboratively train deep learning models by partitioning neural networks, while AL focuses on aggregating intermediate results or model updates from multiple participants, improving robustness, optimizing resource utilization, and mitigating data leakage risks. Specifically, SL is ideal for scenarios requiring strict data isolation (e.g., vertical collaborations), whereas AL suits homogeneous horizontal data settings; they can be combined to balance privacy and communication efficiency. This survey provides a comprehensive analysis of SL and AL in 6G communication systems, exploring their architectures, technical methodologies, and integration with AI-native 6G communication technologies. We examine different SL configurations, aggregation techniques, and their roles in optimizing distributed foundation models. Furthermore, we discuss their applications in emerging wireless networks, including semantic communication, reconfigurable intelligent surfaces (RIS), space-air-ground integrated networks (SAGINs), and quantum communication. By analyzing the impact of SL and AL, this survey provides insights into their role in shaping distributed AI-driven communication systems in the 6G era, focusing on efficiency, privacy preservation, and scalability.

cs.IT↗

arXiv2026

Complexity of Auctions with Interdependence

We study auction design in the celebrated interdependence model introduced by Milgrom and Weber [1982], where a mechanism designer allocates a good, maximizing the value of the agent who receives it, while inducing truthfulness using payments. In the lesser-studied procurement auctions, one allocates a chore, minimizing the cost incurred by the agent selected to perform it. Most of the past literature in theoretical computer science considers designing truthful mechanisms with constant approximation for the value setting, with restricted domains and monotone valuation functions. In this work, we study the general computational problems of optimizing the approximation ratio of truthful mechanism, for both value and cost, in the deterministic and randomized settings. Unlike most previous works, we remove the domain restriction and the monotonicity assumption imposed on value functions. We provide theoretical explanations for why some previously considered special cases are tractable, reducing them to classical combinatorial problems, and providing efficient algorithms and characterizations. We complement our positive results with hardness results for the general case, providing query complexity lower bounds, and proving the NP-Hardness of the general case.

cs.GT↗

arXiv2026

CovertComBench: A First Domain-Specific Testbed for LLMs in Wireless Covert Communication

The integration of Large Language Models (LLMs) into wireless networks presents significant potential for automating system design. However, unlike conventional throughput maximization, Covert Communication (CC) requires optimizing transmission utility under strict detection-theoretic constraints, such as Kullback-Leibler divergence limits. Existing benchmarks primarily focus on general reasoning or standard communication tasks and do not adequately evaluate the ability of LLMs to satisfy these rigorous security constraints. To address this limitation, we introduce CovertComBench, a unified benchmark designed to assess LLM capabilities across the CC pipeline, encompassing conceptual understanding (MCQs), optimization derivation (ODQs), and code generation (CGQs). Furthermore, we analyze the reliability of automated scoring within a detection-theoretic ``LLM-as-Judge'' framework. Extensive evaluations across state-of-the-art models reveal a significant performance discrepancy. While LLMs achieve high accuracy in conceptual identification (81%) and code implementation (83%), their performance in the higher-order mathematical derivations necessary for security guarantees ranges between 18% and 55%. This limitation indicates that current LLMs serve better as implementation assistants rather than autonomous solvers for security-constrained optimization. These findings suggest that future research should focus on external tool augmentation to build trustworthy wireless AI systems.

cs.NI↗

arXiv2025

Hyperion: Hierarchical Scheduling for Parallel LLM Acceleration in Multi-tier Networks

LLMs are increasingly executed in edge where limited GPU memory and heterogeneous computation jointly constrain deployment which motivates model partitioning and request scheduling. In this setting, minimizing latency requires addressing the tight coupling between model placement and request scheduling across heterogeneous nodes, as suboptimal decisions in one domain can negate benefits in the other. In this paper, we propose Hyperion, a hierarchical two-stage framework that jointly optimizes partitioning and scheduling for pipelined LLM inference. Hyperion minimizes latency by balancing resources across tiers without requiring model retraining or incurring significant runtime overhead. Leveraging the timescale difference between partitioning and request arrivals, Stage 1 performs offline, inter-tier partitioning via a Hyperion Split with Dynamic Programming (HypSplit-DP) procedure to produce balanced stage times under tier capacity and memory constraints; to adapt to time-varying load, Stage 2 performs online, intra-tier scheduling with a lightweight Hyperion Scheduling for Real-Time (HypSched-RT) that maps each request to the best available node using real-time estimates of queue length and effective capacity. Experiments with Phi-3-medium demonstrate that Hyperion reduces latency by up to 52.1% (vs. GPipe) and 31.2% (vs. HEFT). Furthermore, Hyperion exhibits superior scalability for long-sequence generation, maintaining 44.5% lower latency and higher GPU utilization.

cs.DC↗

arXiv2025

R&D-Agent: An LLM-Agent Framework Towards Autonomous Data Science

Recent advances in AI and ML have transformed data science, yet increasing complexity and expertise requirements continue to hinder progress. Although crowd-sourcing platforms alleviate some challenges, high-level machine learning engineering (MLE) tasks remain labor-intensive and iterative. We introduce R&D-Agent, a comprehensive, decoupled, and extensible framework that formalizes the MLE process. R&D-Agent defines the MLE workflow into two phases and six components, turning agent design for MLE from ad-hoc craftsmanship into a principled, testable process. Although several existing agents report promising gains on their chosen components, they can mostly be summarized as a partial optimization from our framework's simple baseline. Inspired by human experts, we designed efficient and effective agents within this framework that achieve state-of-the-art performance. Evaluated on MLE-Bench, the agent built on R&D-Agent ranks as the top-performing machine learning engineering agent, achieving 35.1% any medal rate, demonstrating the ability of the framework to speed up innovation and improve accuracy across a wide range of data science applications. We have open-sourced R&D-Agent on GitHub: https://github.com/microsoft/RD-Agent.

cs.AI↗

arXiv2025

R&D-Agent-Quant: A Multi-Agent Framework for Data-Centric Factors and Model Joint Optimization

Financial markets pose fundamental challenges for asset return prediction due to their high dimensionality, non-stationarity, and persistent volatility. Despite advances in large language models and multi-agent systems, current quantitative research pipelines suffer from limited automation, weak interpretability, and fragmented coordination across key components such as factor mining and model innovation. In this paper, we propose R&D-Agent for Quantitative Finance, in short RD-Agent(Q), the first data-centric multi-agent framework designed to automate the full-stack research and development of quantitative strategies via coordinated factor-model co-optimization. RD-Agent(Q) decomposes the quant process into two iterative stages: a Research stage that dynamically sets goal-aligned prompts, formulates hypotheses based on domain priors, and maps them to concrete tasks, and a Development stage that employs a code-generation agent, Co-STEER, to implement task-specific code, which is then executed in real-market backtests. The two stages are connected through a feedback stage that thoroughly evaluates experimental outcomes and informs subsequent iterations, with a multi-armed bandit scheduler for adaptive direction selection. Empirically, RD-Agent(Q) achieves up to 2X higher annualized returns than classical factor libraries using 70% fewer factors, and outperforms state-of-the-art deep time-series models on real markets. Its joint factor-model optimization delivers a strong balance between predictive accuracy and strategy robustness. Our code is available at: https://github.com/microsoft/RD-Agent.

q-fin.CP↗

arXiv2025

A QoE-Driven Personalized Incentive Mechanism Design for AIGC Services in Resource-Constrained Edge Networks

With rapid advancements in large language models (LLMs), AI-generated content (AIGC) has emerged as a key driver of technological innovation and economic transformation. Personalizing AIGC services to meet individual user demands is essential but challenging for AIGC service providers (ASPs) due to the subjective and complex demands of mobile users (MUs), as well as the computational and communication resource constraints faced by ASPs. To tackle these challenges, we first develop a novel multi-dimensional quality-of-experience (QoE) metric. This metric comprehensively evaluates AIGC services by integrating accuracy, token count, and timeliness. We focus on a mobile edge computing (MEC)-enabled AIGC network, consisting of multiple ASPs deploying differentiated AIGC models on edge servers and multiple MUs with heterogeneous QoE requirements requesting AIGC services from ASPs. To incentivize ASPs to provide personalized AIGC services under MEC resource constraints, we propose a QoE-driven incentive mechanism. We formulate the problem as an equilibrium problem with equilibrium constraints (EPEC), where MUs as leaders determine rewards, while ASPs as followers optimize resource allocation. To solve this, we develop a dual-perturbation reward optimization algorithm, reducing the implementation complexity of adaptive pricing. Experimental results demonstrate that our proposed mechanism achieves a reduction of approximately $64.9\%$ in average computational and communication overhead, while the average service cost for MUs and the resource consumption of ASPs decrease by $66.5\%$ and $76.8\%$, respectively, compared to state-of-the-art benchmarks.

cs.GT↗

arXiv2025

Does Learning Mathematical Problem-Solving Generalize to Broader Reasoning?

There has been a growing interest in enhancing the mathematical problem-solving (MPS) capabilities of large language models. While the majority of research efforts concentrate on creating specialized models to solve mathematical problems, it remains unknown how learning mathematical problem-solving generalizes to help develop other reasoning abilities. In this paper, we present an empirical investigation into the generalization potential of various MPS training approaches, such as continual pretraining, instruction tuning, and rule-based reinforcement learning across various data sources, including both short and long chain-of-thought (CoT) samples. Evaluation on 5 mathematical and 8 general reasoning benchmarks show that continual pretraining on math text is able to generalize to general reasoning tasks to some extent. In constrast, instruction tuning on conventional, short MPS samples provides limited benefits and, in many cases, even impairs generalization performance. Notably, training with long CoT responses for MPS samples and incorporating rule-based reinforcement learning on MPS queries exhibit distinct behavior, significantly enhancing generalization by extending the model's reasoning processes into other domains. These results suggest that traditional approaches to learning MPS with short reasoning chains largely fail to achieve robust generalization. However, the emerging paradigm of longer reasoning chains, coupled with self-reflection, offers a promising direction for improving generalized reasoning abilities through learning from specialized domains.

cs.CL↗

arXiv2025

Forewarned is Forearmed: A Survey on Large Language Model-based Agents in Autonomous Cyberattacks

With the continuous evolution of Large Language Models (LLMs), LLM-based agents have advanced beyond passive chatbots to become autonomous cyber entities capable of performing complex tasks, including web browsing, malicious code and deceptive content generation, and decision-making. By significantly reducing the time, expertise, and resources, AI-assisted cyberattacks orchestrated by LLM-based agents have led to a phenomenon termed Cyber Threat Inflation, characterized by a significant reduction in attack costs and a tremendous increase in attack scale. To provide actionable defensive insights, in this survey, we focus on the potential cyber threats posed by LLM-based agents across diverse network systems. Firstly, we present the capabilities of LLM-based cyberattack agents, which include executing autonomous attack strategies, comprising scouting, memory, reasoning, and action, and facilitating collaborative operations with other agents or human operators. Building on these capabilities, we examine common cyberattacks initiated by LLM-based agents and compare their effectiveness across different types of networks, including static, mobile, and infrastructure-free paradigms. Moreover, we analyze threat bottlenecks of LLM-based agents across different network infrastructures and review their defense methods. Due to operational imbalances, existing defense methods are inadequate against autonomous cyberattacks. Finally, we outline future research directions and potential defensive strategies for legacy network systems.

cs.NI↗

arXiv2025

ScaleBiO: Scalable Bilevel Optimization for LLM Data Reweighting

Bilevel optimization has shown its utility across various machine learning settings, yet most algorithms in practice require second-order information, making it challenging to scale them up. Only recently, a paradigm of first-order algorithms has emerged in the theoretical literature, capable of effectively addressing bilevel optimization problems. Nevertheless, the practical efficiency of this paradigm remains unverified, particularly in the context of large language models (LLMs). This paper introduces the first scalable instantiation of this paradigm called ScaleBiO, focusing on bilevel optimization for large-scale LLM data reweighting. By combining with a recently proposed memory-efficient training technique called LISA, our novel algorithm allows the paradigm to scale to $\sim$30B-sized LLMs on $8\times$H100 GPUs, marking the first successful application of bilevel optimization under practical scenarios for large-sized LLMs. Empirically, extensive experiments on data reweighting verify the effectiveness of ScaleBiO for different-scaled models, including Llama-3-8B, Gemma-2-9B, Qwen-2-7B, and Qwen-2.5-32B, where bilevel optimization succeeds in instruction-following and math reasoning tasks, outperforming several popular baselines, including uniform sampling, influence-aware data filtering, and reference-model-based sampling methods. Theoretically, ScaleBiO ensures the optimality of the learned data weights, along with a convergence guarantee matching the conventional first-order bilevel optimization paradigm on smooth and strongly convex objectives.

cs.LG↗

arXiv2025

Shadow Wireless Intelligence: Large Language Model-Driven Reasoning in Covert Communications

Covert Communications (CC) can secure sensitive transmissions in industrial, military, and mission-critical applications within 6G wireless networks. However, traditional optimization methods based on Artificial Noise (AN), power control, and channel manipulation might not adapt to dynamic and adversarial environments due to the high dimensionality, nonlinearity, and stringent real-time covertness requirements. To bridge this gap, we introduce Shadow Wireless Intelligence (SWI), which integrates the reasoning capabilities of Large Language Models (LLMs) with retrieval-augmented generation to enable intelligent decision-making in covert wireless systems. Specifically, we utilize DeepSeek-R1, a mixture-of-experts-based LLM with RL-enhanced reasoning, combined with real-time retrieval of domain-specific knowledge to improve context accuracy and mitigate hallucinations. Our approach develops a structured CC knowledge base, supports context-aware retrieval, and performs semantic optimization, allowing LLMs to generate and adapt CC strategies in real time. In a case study on optimizing AN power in a full-duplex CC scenario, DeepSeek-R1 achieves 85% symbolic derivation accuracy and 94% correctness in the generation of simulation code, outperforming baseline models. These results validate SWI as a robust, interpretable, and adaptive foundation for LLM-driven intelligent covert wireless systems in 6G networks.

cs.NI↗

arXiv2025

Movable Antenna-Aided Federated Learning with Over-the-Air Aggregation: Joint Optimization of Positioning, Beamforming, and User Selection

Federated learning (FL) in wireless computing effectively utilizes communication bandwidth, yet it is vulnerable to errors during the analog aggregation process. While removing users with unfavorable channel conditions can mitigate these errors, it also reduces the available local training data for FL, which in turn hinders the convergence rate of the training process. To tackle this issue, we propose the use of movable antenna (MA) techniques to enhance the degrees of freedom within the channel space, ultimately boosting the convergence speed of FL training. Moreover, we develop a coordinated approach for uplink receiver beamforming, user selection, and MA positioning to optimize the convergence rate of wireless FL training in dynamic wireless environments. This stochastic optimization challenge is reformulated into a mixed-integer programming problem by utilizing the training loss upper bound. We then introduce a penalty dual decomposition (PDD) method to solve the mixed-integer mixed programming problem. Experimental results indicate that incorporating MA techniques significantly accelerates the training convergence of FL and greatly surpasses conventional methods.

eess.SP↗

arXiv2025

Hybrid Reinforcement Learning-based Sustainable Multi-User Computation Offloading for Mobile Edge-Quantum Computing

Exploiting quantum computing at the mobile edge holds immense potential for facilitating large-scale network design, processing multimodal data, optimizing resource management, and enhancing network security. In this paper, we propose a pioneering paradigm of mobile edge quantum computing (MEQC) that integrates quantum computing capabilities into classical edge computing servers that are proximate to mobile devices. To conceptualize the MEQC, we first design an MEQC system, where mobile devices can offload classical and quantum computation tasks to edge servers equipped with classical and quantum computers. We then formulate the hybrid classical-quantum computation offloading problem whose goal is to minimize system cost in terms of latency and energy consumption. To solve the offloading problem efficiently, we propose a hybrid discrete-continuous multi-agent reinforcement learning algorithm to learn long-term sustainable offloading and partitioning strategies. Finally, numerical results demonstrate that the proposed algorithm can reduce the MEQC system cost by up to 30% compared to existing baselines.

cs.NI↗

arXiv2025

Efficient Twin Migration in Vehicular Metaverses: Multi-Agent Split Deep Reinforcement Learning with Spatio-Temporal Trajectory Generation

Vehicle Twins (VTs) as digital representations of vehicles can provide users with immersive experiences in vehicular metaverse applications, e.g., Augmented Reality (AR) navigation and embodied intelligence. VT migration is an effective way that migrates the VT when the locations of physical entities keep changing to maintain seamless immersive VT services. However, an efficient VT migration is challenging due to the rapid movement of vehicles, dynamic workloads of Roadside Units (RSUs), and heterogeneous resources of the RSUs. To achieve efficient migration decisions and a minimum latency for the VT migration, we propose a multi-agent split Deep Reinforcement Learning (DRL) framework combined with spatio-temporal trajectory generation. In this framework, multiple split DRL agents utilize split architecture to efficiently determine VT migration decisions. Furthermore, we propose a spatio-temporal trajectory generation algorithm based on trajectory datasets and road network data to simulate vehicle trajectories, enhancing the generalization of the proposed scheme for managing VT migration in dynamic network environments. Finally, experimental results demonstrate that the proposed scheme not only enhances the Quality of Experience (QoE) by 29% but also reduces the computational parameter count by approximately 25% while maintaining similar performances, enhancing users' immersive experiences in vehicular metaverses.

cs.NI↗

arXiv2025

Adaptive AUV Hunting Policy with Covert Communication via Diffusion Model

Collaborative underwater target hunting, facilitated by multiple autonomous underwater vehicles (AUVs), plays a significant role in various domains, especially military missions. Existing research predominantly focuses on designing efficient and high-success-rate hunting policy, particularly addressing the target's evasion capabilities. However, in real-world scenarios, the target can not only adjust its evasion policy based on its observations and predictions but also possess eavesdropping capabilities. If communication among hunter AUVs, such as hunting policy exchanges, is intercepted by the target, it can adapt its escape policy accordingly, significantly reducing the success rate of the hunting mission. To address this challenge, we propose a covert communication-guaranteed collaborative target hunting framework, which ensures efficient hunting in complex underwater environments while defending against the target's eavesdropping. To the best of our knowledge, this is the first study to incorporate the confidentiality of inter-agent communication into the design of target hunting policy. Furthermore, given the complexity of coordinating multiple AUVs in dynamic and unpredictable environments, we propose an adaptive multi-agent diffusion policy (AMADP), which incorporates the strong generative ability of diffusion models into the multi-agent reinforcement learning (MARL) algorithm. Experimental results demonstrate that AMADP achieves faster convergence and higher hunting success rates while maintaining covertness constraints.

cs.MA↗

arXiv2025

Fluid Antenna Enabled Over-the-Air Federated Learning: Joint Optimization of Positioning, Beamforming, and User Selection

Over-the-air (OTA) federated learning (FL) effectively utilizes communication bandwidth, yet it is vulnerable to errors during analog aggregation. While removing users with unfavorable channel conditions can mitigate these errors, it also reduces the available local training data for FL, which in turn hinders the convergence rate of the training process. To tackle this issue, we propose using fluid antenna (FA) techniques to enhance the degrees of freedom within the channel space, ultimately boosting the convergence speed of FL training. Moreover, we develop a novel approach that effectively coordinates uplink receiver beamforming, user selection, and FA positioning to optimize the convergence rate of OTA FL training in dynamic wireless environments. We address this challenging stochastic optimization by reformulating it as a mixed-integer programming problem by utilizing the training loss upper bound. We then introduce a penalty dual decomposition (PDD) method to solve the mixed-integer mixed programming problem. Experimental results indicate that incorporating FA techniques significantly accelerates the training convergence of FL and greatly surpasses conventional methods.

eess.SP↗