Searcharxiv⌕ Search

arXiv subjects

Feng Xiao

Publications and source records attributed to Feng Xiao.

At least 19 recordsLinked to original sources
arXiv2026

IRPM: Intergroup Relative Preference Modeling for Pointwise Generative Reward Models

Generative Reward Models (GRMs) have demonstrated strong performance in reward modeling, due to their interpretability and potential for refinement through reinforcement learning (RL). However, widely used pairwise GRMs create a computational bottleneck in reinforcement learning from human feedback (RLHF), when calibrating or aggregating preference signals over n candidates, often incurring O(n^2) pairwise judgments. To address this issue, we propose Intergroup Relative Preference Modeling (IRPM), an RL-based method that extends the Bradley-Terry preference-learning paradigm via intergroup comparisons to train pointwise GRMs from pairwise preference data. IRPM derives a pointwise reward for each response by contrasting groups of chosen vs. rejected samples, enabling pointwise scores comparable across candidate sets and O(n) reward evaluation for a variable number of candidates during RL training, while preserving interpretability and scalability. Experiments show that IRPM achieves state-of-the-art performance among pointwise GRMs on JudgeBench and RewardBench, and approaches the performance of leading pairwise GRMs. In addition, IRPM achieves substantial gains in post-training evaluations, demonstrating its effectiveness. Our models and code are available at https://github.com/XINGYUN-AI-LAB/IRPM.

cs.LG↗
arXiv2026

Stability and Optimal Design of Adaptive Linear Bus Holding on Open Chains and Loops

Bus holding can suppress bunching, but its stability and performance depend on the feedback structure and on how the bus sequence is bounded. This paper develops a common analytical framework for adaptive linear holding on finite open chains and loops and relates both to the bi-infinite system studied in the literature. Matrix and Fourier methods yield necessary and sufficient stability conditions for current- and leading-bus kernels on open chains and general cyclic kernels on loops, together with computable covariance bounds on open chains and exact variance formulas on loops. The analysis shows when periodic closure changes stability, when coefficient-sum stability tests in the literature are conservative, and why a stable open chain can still accumulate substantial variance before boundary effects arrive. For loops with same-lap feedback, modal analysis links control coefficients to fleet-wide schedule drift and to relative headway patterns, explaining the behavior of existing rules and yielding closed-form Pareto-optimal designs for schedule-holding and headway-holding tradeoffs. Cross-lap output corrections relate these designs to physical headway and holding performance. Numerical experiments show that the analytical design and a three-tap approximation, both obtained from closed-form coefficients and a one-dimensional search, outperform four established controls at the tested moderate and loose headway-variance budgets, with holding variances within 0.35% and 3.5% of the full cyclic numerical frontier across these budgets and other parameter variations. The framework thus provides a tractable design procedure and an explanation of how feedback structure and boundary conditions shape service reliability and holding effort.

physics.app-ph↗
arXiv2026

From Observation to Intervention: Mastering Bus Holding Control from Historical Uncontrolled Trajectories with Offline Reinforcement Learning

Bus operations are inherently unstable---buses initially dispatched at equal headways will inevitably end up bunching up. This issue exacerbates passenger waiting times, creates load imbalances between buses, and undermines service reliability. Reinforcement learning (RL) has proven highly effective for mitigating bus bunching and outperforms traditional model-based approaches. However, existing RL approaches require agents to continuously interact with the environment, which potentially disrupts the bus system before effective policies are learned. This paper proposes an Offline Graph-based Multi-agent Actor-Critic (Offline-GMAC) method that trains bus holding policies using only readily available logged GPS data collected during normal uncontrolled operations, thereby eliminating the need for disruptive online interaction or exploration. The method employs general graph neural networks for multi-agent credit assignment. A key innovation of our method is a data augmentation scheme that synthesizes unperformed holding actions from historical uncontrolled bus trajectories. The scheme exploits the equivalence between stochastic link travel time delays and deliberate holding interventions. Under mild assumptions, we prove that the 1-Wasserstein distance between synthesized and target state transition distributions is bounded above by a term that scales linearly with the standard deviation of link travel times plus a tunable synthesis-dependent term. Evaluations using real-world bus data demonstrate that the proposed offline method achieves performance comparable to, and in some cases, even better than its online counterpart.

econ.GN↗
arXiv2026

Distributed Nash Equilibrium Seeking for Open Multi-Coalition Games: A Mass-Preserving Gradient-Tracking-Based Algorithm

This paper studies distributed information aggregation and Nash equilibrium seeking over open multi-coalition networks, where both agents and coalitions may dynamically enter and leave the network. The varying network population poses a fundamental challenge to conventional gradient-tracking methods, as agent departures may cause the loss of gradient information required for distributed aggregation. To overcome this issue, a mass-preserving gradient-tracking method is designed, in which the gradient information carried by departing agents is redistributed to the remaining agents within the same coalition. The proposed method preserves the aggregate gradient information of the coalition objective over the intra-coalition communication network and achieves gradient tracking under dynamic population changes without imposing additional initialization conditions beyond those of conventional gradient-tracking methods. Based on the proposed gradient-tracking mechanism, a distributed Nash equilibrium seeking framework is developed, and its convergence is established under both uncoordinated fixed and diminishing stepsizes. Furthermore, to alleviate the slowdown of convergence caused by classical diminishing stepsize schemes ($\frac{η_0}{t}$, where $η_0>0$ is a constant), an exponential-interval diminishing stepsize scheme is designed based on the agents' exploration intervals. Finally, a numerical simulation based on an open connecting control game demonstrate the effectiveness of the proposed information-tracking and Nash equilibrium seeking framework.

eess.SY↗
arXiv2026

The Next Challenge for Agentic Cybersecurity: A Realistic, Contamination-Free Reverse Engineering Benchmark

AI agents are rapidly improving in cybersecurity when source code is available, yet much of the software most consequential to security, including malware, firmware, and proprietary applications, exists only as binaries. Analyzing such software requires reverse engineering (RE): recovering program semantics before analysis can proceed. Evaluating agentic RE poses a fundamental challenge: realistic benchmark instances must (1) be absent from LLMs' training data to prevent shortcuts by memorization, and (2) reflect the scale and anti-analysis protections of real-world binaries. We introduce SRE-Bench, the first realistic, contamination-free RE benchmark. Built from scratch by RE experts with over 5,000 expert hours, SRE-Bench comprises 19 private, real-world-scale programs averaging 16.9K lines of code. We further developed 44 in-house anti-analysis mechanisms, yielding 262 binary instances and 1,572 deterministically graded tasks. We evaluated 13 agentic settings across 11 models: eight in public-facing settings and five in internal unconstrained settings with cyber safeguards disabled and no budget cap. Realistic RE remains challenging for frontier agents: GPT-5.6-Sol and Claude-Fable-5.1, despite strong source-code security capabilities, fully solve only 31.5% and 26.9% of graded instances, suggesting that success in source-code security does not translate into effective binary analysis. Without a budget cap and safety guard, GPT-6-Astra achieves a near-perfect pass@4 score, yet reliably identifying the correct candidate remains difficult. Agents are largely insensitive to compiler optimization and static linking, and ablations confirm that both contamination control and realistic scale are essential to understanding agents' RE capability. These findings highlight RE as a distinct frontier for agentic cybersecurity and establish SRE-Bench as a rigorous testbed for measuring progress.

cs.CR↗
arXiv2026

Towards On-Device Evidence Gathering for Intimate Partner Infiltration: A Feasibility Study for Joint Identity-Action Detection

Intimate Partner Infiltration (IPI) refers to phone-side privacy infiltration in intimate or close relationships, often enabled by physical access to a person's smartphone and discussed in technology-facilitated Intimate Partner Violence (IPV) contexts. Unlike conventional cyberattackers, IPI perpetrators leverage proximity and personal knowledge to circumvent standard protection, underscoring the need for targeted interventions, motivating device-side tools that surface such risk evidence for later review. While prior works have extensively studied IPV, and some have provided tailored and effective solutions such as security clinics, they are necessarily episodic and human-expert-intensive, and offer limited automated visibility into what happens on a smartphone between support sessions. Guided by a formative interview with experts (n=5), we take the first exploration into gathering IPI-risk evidence from a mobile system perspective and present AID, Automated IPI Detection, a data-driven system that continuously logs unauthorized access and suspicious behaviors on smartphones. In a controlled 27-participant study, AID achieves an end-to-end F1 score of 0.928 with a 7.0% false positive rate for Top-1 phone-side risk flagging; when preserving top-3 candidate action categories as report context, AID achieves an F1 score of 0.981 and a false positive rate of 1.6%. These findings demonstrate AID's potential as an evidence-support tool that complements current clinic-based interpretation and safety-planning.

cs.CR↗
arXiv2026

Text-ADBench: Text Anomaly Detection Benchmark Based on LLM Embeddings

Text anomaly detection is a critical task in natural language processing (NLP), with applications spanning fraud detection, misinformation identification, spam detection and content moderation, etc. Despite significant advances in large language models (LLMs) and anomaly detection algorithms, the absence of standardized and comprehensive benchmarks for evaluating the existing anomaly detection methods on text data limits rigorous comparison and development of innovative approaches. This work performs a comprehensive empirical study and introduces a benchmark for text anomaly detection, leveraging embeddings from diverse pre-trained language models across a wide array of text datasets. Our work systematically evaluates the effectiveness of embedding-based text anomaly detection by incorporating (1) early language models (GloVe, BERT); (2) multiple LLMs (LLaMA-2, LLaMA-3, Mistral, OpenAI embedding models (small, ada, large)); (3) multi-domain text datasets (news, social media, scientific publications); (4) comprehensive evaluation metrics (AUROC, AUPRC). Our experiments reveal a critical empirical insight: embedding quality significantly governs anomaly detection efficacy, and deep learning-based approaches demonstrate no performance advantage over conventional shallow algorithms (e.g., KNN, OCSVM) when leveraging LLM-derived embeddings. In addition, we observe strongly low-rank characteristics in cross-model performance matrices, which enables an efficient strategy for rapid model evaluation (or embedding evaluation) and selection in practical applications. Furthermore, by open-sourcing our benchmark toolkit that includes all embeddings from different models and code, this work provides a foundation for future research in robust and scalable text anomaly detection systems. Our code repository is at https://github.com/jicongfan/Text-Anomaly-Detection-Benchmark.

cs.CL↗
arXiv2026

Step-Level On-Policy Distillation: Interpolating Between On-Policy Distillation and Supervised Fine-Tuning

On-policy distillation (OPD) aligns a student model with a teacher's logit distribution on student-generated trajectories. This approach has achieved strong empirical gains and can often surpass conventional off-policy distillation with substantially less data. However, standard token-level OPD can provide only fragmented corrections along an erroneous student trajectory and cannot unfold a complete and correct repair path. Motivated by this limitation, we propose \emph{Step-Level On-Policy Distillation} (SOPD), which combines the long-horizon correction of supervised fine-tuning (SFT) with the on-policy advantage of OPD to provide step-level supervision over complete student-generated trajectories. We show that, at different limits of step length, SOPD reduces to SFT or approximates OPD. Compared with SFT, the teacher responses in SOPD are conditioned on student trajectories and therefore align more closely with student-visited states; compared with OPD, SOPD provides longer-horizon corrections rather than fragmented token-level guidance. Across both reasoning and agent tasks, SOPD substantially outperforms conventional SFT and OPD. For example, on ALFWorld, SOPD improves the average success rate by 13.4 points over Vanilla OPD. We hope this work offers a new perspective for future research on distillation methods.

cs.CL↗
arXiv2026

MDLMPE: Distribution Aware Positional Encoding for Masked Diffusion Language Models

Masked diffusion language models (MDLMs) enable parallel generation and bidirectional context modeling, but their positional context differs fundamentally from that of autoregressive (AR) models. Whereas AR decoding exposes a contiguous prefix, MDLM denoising produces dynamic, non-contiguous configurations of revealed and masked tokens. Conventional positional encodings such as RoPE capture sequence order and pairwise displacement but remain insensitive to this evolving token-availability structure. To address this limitation, we propose MDLMPE, a positional encoding designed specifically for masked diffusion. To the best of our knowledge, MDLMPE is the first method to make positional representations explicitly aware of the changing revealed/masked configuration. It represents token availability as a binary sequence, applies distance-aware Gaussian weighting, and projects the resulting pattern through a cosine basis to obtain distribution-aware positional features. These features are added to token embeddings and mapped by a lightweight MLP to angular offsets that modulate the standard RoPE phases. Extensive experiments on LLaDA and DREAM demonstrate that MDLMPE generally outperforms conventional positional encoding methods across supervised fine-tuning, pretraining, zero-shot evaluation, and block-diffusion settings. Further ablations show that the complete combination of availability state, Gaussian locality, spectral basis, and embedding injection yields the strongest result. These results establish the evolving token-availability distribution as a useful positional signal for masked diffusion language models.

cs.CL↗
arXiv2026

Separating Capability from Permission: A Governance Framework for Agentic AI Autonomy Levels

As AI systems increasingly exhibit agentic behavior, discussions of autonomy often conflate what systems are technically capable of doing with what they should be permitted to do in practice. This paper introduces a governance framework that explicitly separates Allowed Autonomy Levels (AAL), which define the degree of autonomy an AI agent is authorized to exercise given risk, oversight, and accountability considerations, from Autonomous Capability Levels (ACL), which characterize an agent's inherent technical abilities. We present a structured set of autonomy levels spanning reactive execution, decision support, supervised action, goal-directed autonomy, and delegated operational authority, and describe how control, reversibility, and accountability change as autonomy increases. To operationalize this framework, we propose a risk-aware decision process for assigning allowed autonomy, analyze how risk and accountability evolve across autonomy levels, and demonstrate its application through a deployed enterprise data engineering agent, illustrating how a system assessed at a high capability level can be deliberately constrained to a lower allowed autonomy based on risk, reversibility, and organizational readiness. By distinguishing authorization from capability, this work provides practical guidance for the design, deployment, and governance of Agentic AI systems.

cs.AI↗
arXiv2026

A Fourth-order Conservative Adaptive Multiresolution Wavelet Upwind Scheme for Compressible Flows

A fourth-order conservative adaptive multiresolution average-interpolating wavelet upwind scheme is proposed for compressible flows governed by hyperbolic conservation laws. A family of asymmetric average-interpolating wavelets with upwind properties is constructed for conservative finite volume discretization, while symmetric average-interpolating wavelets are employed for multiresolution decomposition and reconstruction of physical variables in the adaptive procedure. Since both the conservative discretization and the adaptive multiresolution representation are constructed from cell-average quantities, the proposed scheme preserves strict conservation during both numerical evolution and adaptive cell redistribution. Unlike hybrid adaptive wavelet methods that use wavelets mainly for data compression and mesh adaptation, the present adaptive wavelet upwind scheme utilizes average-interpolating wavelet multiresolution approximation to reconstruct the interface values directly for numerical flux evaluation, thereby avoiding additional ghost-cell marking and reconstruction near coarse--fine mesh interfaces. The boundary variation diminishing reconstruction is incorporated at the finest resolution level to achieve non-oscillatory shock-capturing capability. Numerical tests demonstrate that the proposed scheme achieves the expected fourth-order accuracy, maintains conservation errors close to machine precision, and controls numerical errors around the prescribed threshold. The proposed method also sharply captures shock waves and contact discontinuities without spurious oscillations and resolves multiscale smooth structures through a sparse adaptive representation. These results indicate that the proposed scheme provides an efficient, conservative, and reliable approach for high-resolution simulations of compressible flows.

math.NA↗
arXiv2026

Outcome-Grounded Advantage Reshaping for Fine-Grained Credit Assignment in Mathematical Reasoning

Group Relative Policy Optimization (GRPO) has emerged as a promising critic-free reinforcement learning paradigm for reasoning tasks. However, standard GRPO employs a coarse-grained credit assignment mechanism that propagates group-level rewards uniformly to to every token in a sequence, neglecting the varying contribution of individual reasoning steps. We address this limitation by introducing Outcome-grounded Advantage Reshaping (OAR), a fine-grained credit assignment mechanism that redistributes advantages based on how much each token influences the model's final answer. We instantiate OAR via two complementary strategies: (1) OAR-P, which estimates outcome sensitivity through counterfactual token perturbations, serving as a high-fidelity attribution signal; (2) OAR-G, which uses an input-gradient sensitivity proxy to approximate the influence signal with a single backward pass. These importance signals are integrated with a conservative Bi-Level advantage reshaping scheme that suppresses low-impact tokens and boosts pivotal ones while preserving the overall advantage mass. Empirical results on extensive mathematical reasoning benchmarks demonstrate that while OAR-P sets the performance upper bound, OAR-G achieves comparable gains with negligible computational overhead, both significantly outperforming a strong GRPO baseline, pushing the boundaries of critic-free LLM reasoning.

cs.CL↗
arXiv2026

Robustness Enhancement of Consensus Networks: the Optimal Memory Depth

Understanding what governs collective robustness and how it can be enhanced remains a central pursuit in network science. This paper investigates the robustness of multi-agent consensus networks, quantified by the $H_2$ performance metric, and delves into the enhancing effect of agents' local memory on it. Inspired by the hierarchical temporal structure of memory observed in neuroscience, we focus on the role of memory depth, which reflects the temporal features of memory from recent to remote. Building on linear extrapolation, we propose a consensus protocol with single-step memory and tunable memory depth, derive the necessary and sufficient condition for achieving consensus, and show that the protocol exhibits an inheritable consensus property across memory depths. Furthermore, analytical expressions for the $H_2$ performance metric, which depend on the memory factor, memory depth, coupling gain, and Laplacian spectrum, are established. Under balanced usage of real-time and memory information, we demonstrate that memory at any accessible depth enhances $H_2$ performance, and the optimal memory depth occurs at either the most recent or the most remote memory, contingent upon certain parameter regions. Further detailed discussions are provided to clarify the broader implications of our findings.

eess.SY↗
arXiv2026

CLUBench: A Clustering Benchmark

Clustering is a fundamental problem in data science with a long-standing research history, yielding numerous insightful algorithms. Despite this progress, a systematic and large-scale empirical evaluation that jointly considers conventional algorithms, deep learning-based methods, and recent foundation model-based clustering remains largely absent, leading to limited guidance on algorithm selection and deployment. To address this gap, we introduce CLUBench, a comprehensive clustering benchmark comprising 24 algorithms of diverse principles evaluated on 131 datasets across tabular, text, and image data, involving 178,815 experiments. Importantly, our analyses of (i) the impact of hyperparameter tuning,(ii) the impact of data types and characteristics,(iii) the impact of pretrained embeddings,(iv) large language model-based clustering,(v) the similarity of algorithms, and (vi) the low-rank structures of performance matrices, yield meaningful insights and promising pathways for clustering research. For instance, our study reveals that: 1) All evaluated deep clustering methods do not exhibit a significant advantage compared with the top-performing conventional clustering algorithms (e.g., KMeans, SpeClu) in terms of average performance; 2) For image and text clustering tasks, combining pretrained embeddings with conventional clustering algorithms (e.g., KMeans, SpeClu) offers effective and efficient clustering; 3) Clustering remains a challenging and nontrivial problem, even in the era of increasingly dominant foundation models. Moreover, we propose to use the low-rank structure in cross-model performance matrices to efficiently approximate the overall performance evaluation in practical applications. We further demonstrate the feasibility of model selection based on the performance matrices across all hyperparameter configurations.

cs.LG↗
arXiv2026

Resolving Long-Tail Ambiguity in Unsupervised 3D Point Cloud Segmentation with Language Priors

Existing approaches for unsupervised 3D point cloud segmentation predominantly rely on a purely visual similarity-based learning-by-clustering paradigm, which suffers from a fundamental limitation: long-tail ambiguity. In such a paradigm, features of minor classes are consistently absorbed by dominant clusters, leading to severely imbalanced predictions. To address this issue, we propose LangTail, a language-guided hierarchical learning framework that leverages the balanced world knowledge encoded in language models to mitigate long-tail ambiguity in unsupervised 3D segmentation. The key idea is to establish multi-level associations between language-derived semantic priors and visually underrepresented minor classes, thereby compensating for the biased attention of purely visual clustering toward dominant classes. Specifically, LangTail first constructs an entity-level semantic prior from language models, capturing balanced and fine-grained world knowledge across categories. These priors are injected into a hierarchical clustering framework via contrastive alignment. This guides multi-granularity semantic structure formation and prevents minor classes from being absorbed by dominant clusters, yielding more discriminative representations for underrepresented categories. Extensive experiments on ScanNet-v2, S3DIS, and nuScenes demonstrate that LangTail consistently outperforms existing methods by significant margins, \ie, +13.5, +12.9, and +8.9 mIoU, respectively. These results demonstrate the effectiveness of language priors in improving the representation of minority classes in 3D point clouds. The code will be released at: https://github.com/Whisky0129/langtail_official.

cs.CV↗
arXiv2026

Triviality Corrected Endogenous Reward

Reinforcement learning for open-ended text generation is constrained by the lack of verifiable rewards, necessitating reliance on judge models that require either annotated data or powerful closed-source models. Inspired by recent work on unsupervised reinforcement learning for mathematical reasoning using confidence-based endogenous rewards, we investigate whether this principle can be adapted to open-ended writing tasks. We find that directly applying confidence rewards leads to Triviality Bias: the policy collapses toward high-probability outputs, reducing diversity and meaningful content. We propose TCER (Triviality Corrected Endogenous Reward), which addresses this bias by rewarding the relative information gain between a specialist policy and a generalist reference policy, modulated by a probability-dependent correction mechanism. Across multiple writing benchmarks and model architectures, TCER achieves consistent improvements without external supervision. Furthermore, TCER also transfers effectively to mathematical reasoning, validating the generality of our approach across different generation tasks.

cs.CL↗
arXiv2026

Dissipative Control of General Linear Time-Delay Systems: Applications of the Kronecker-Seuret Decomposition

Stabilizing autonomous linear time delay systems, particularly when addressing an unlimited number of pointwise and distributed delays (DDs) under dissipative constraints, poses a significant challenge. Existing solutions are often hindered by theoretical limitations, numerical obstacles, or an inability to address the complexities of the delay integral kernels. In this paper, we propose a unified framework to tackle the above problem by employing the concept of the Kronecker-Seuret decomposition (KSD) for matrix-valued functions, which we recently have developed for the analysis of complex delay structures in coordination with the Krasovski\uı functional approach. Our strategy can simultaneously address two distinct control problems, where the matrix kernels of DDs can contain an unlimited number of square-integrable functions. We show in detail how the KSD can factorize and approximate different kernel functions simultaneously without introducing conservatism. Furthermore, the use of KSD also enables us to construct complete-type functionals, whose integral kernels can include any number of weakly differentiable and linearly independent functions, underpinned by the utilization of novel integral inequalities derived from the least-squares principle. The solution to each synthesis problem comprises two theorems accompanied by an iterative algorithm, which can be utilized as a single package to compute controller gains, thus eliminating the need for nonlinear solvers. We present the testing results of two challenging examples, which could not be addressed by existing methods, to demonstrate the effectiveness of our methodology. Additionally, the paper reviews recent advancements in the research of time-delay systems, providing a valuable reference for both emerging and established researchers.

math.OC↗
arXiv2026

Robust Control of General Linear Delay Systems under Dissipativity: Part I -- A KSD-based Framework

This paper introduces an effective framework for designing memoryless dissipative full-state feedback for general linear delay systems via the Krasovskiĭ functional (KF) approach, where an arbitrary finite number of pointwise and general distributed delays (DDs) exists in the state, input and output. To handle the infinite dimensionality of DDs, we employ the Kronecker-Seuret Decomposition (KSD) which we recently proposed for analyzing matrix-valued functions in the context of delay systems. The KSD enables factorization or least-squares approximation of any number of $\fL^2$ DD kernels from any number of DDs without introducing conservatism. This also facilitates the construction of a complete-type KF with flexible integral kernels by means of a novel integral inequality derived from the least-squares principle. Our solution includes two theorems and an iterative algorithm to compute controller gains without relying on nonlinear solvers. A numerical example is tested to show the effectiveness of the proposed approach.

math.OC↗