arXiv subjects

Zhaoxi Zhang

Publications and source records attributed to Zhaoxi Zhang.

At least 19 recordsLinked to original sources

arXiv2026

Designing, Deployment and Field Testing of C2Stack for Networked Intelligent Software-Defined UAVs

Unmanned Aerial Vehicles (UAVs) are emerging as critical enablers of next-generation wireless networking and autonomous systems. Despite their potential, deploying and testing networked UAV systems in real-world environments remains challenging, largely due to the absence of well-developed, end-to-end, ready-to-use protocol stacks. To fill this gap, we present C2Stack, a configurable protocol stack and experimental framework designed for real-time control, evaluation, and optimization of UAV networks. C2Stack incorporates a modular control plane, referred to as the~C2Stack Network Operating System (CNOS), alongside a programmable data plane that exposes APIs for cross-layer algorithm development, digital twin integration, and autonomous swarm control. In this article, we share our experience with the deployment and testing of C2Stack. We implemented C2Stack on a custom UAV swarm platform that integrates multiprocessor system-on-chip (MPSoC) radios with Intel NUC computing modules, enabling interoperability with various RF front ends. Field trials were conducted in both netted environments and large-scale outdoor test ranges, focusing on two representative use cases: (i) network utility maximization through online reinforcement learning, and (ii) collaborative interference source localization. The experiments demonstrate the feasibility of real-time, data-driven optimization in dynamic aerial environments, while also revealing practical challenges in field deployments of networked UAV systems, including power constraints, sensing limitations, and deployment logistics. We have made C2Stack source code available to the community under the MIT License, with the goal of establishing it as a foundational framework for experimental research on intelligent networked aerial systems.

eess.SY↗

arXiv2026

Are You Still the Agent I Authorized? Earned Authority under a Fixed Ceiling for Evolving Agents

Long-lived AI agents increasingly evolve after deployment by retaining experience, acquiring skills and tools, revising workflows, delegating work, and moving across task phases. This improves adaptation but creates a distinct authorization problem. Tool-enabled agents can turn model errors and prompt injections into consequential external actions; when evolution occurs under a live grant, the subject exercising that authority or the context in which it acts may no longer match what the user evaluated. Evolution can change both the effects reachable under an old grant and the authority required by the task, which may rise, fall, or become incomparable. Existing tool policies constrain actions but do not determine when a grant survives this change. We formulate authorization continuity: when does an existing grant remain valid, how may active authority change, and what boundary must never move? Our state-bound model fixes a transition envelope and an immutable effect ceiling at grant time. The envelope determines whether the grant survives a mutation; below the ceiling, authority may contract freely and expand only under specified evidence conditions. We distinguish requested from realized effects and prove that, under complete mediation, sound effect abstraction, attenuating delegation, and monitor integrity, mutation cannot amplify protected effects beyond the user-issued ceiling. Agent-produced evidence may allocate authority below the ceiling but cannot raise it. Finally, we map six mutation classes to their authorization consequences.

cs.AI↗

arXiv2026

The generalized underlap coefficient with an application in clustering

Quantifying distributional separation across groups is fundamental in statistical learning and scientific discovery, yet most classical discrepancy measures are tailored to two-group comparisons. We generalize the underlap coefficient (UNL), a multi-group separation measure, to multivariate settings. We study its relationship with Bayes risk and mutual information, and further interpret the UNL as a measure of dependence between group labels and variables of interest. We propose an efficient importance sampling estimator of the UNL that can be combined with flexible density estimation methods. A key application is the assessment of partition-covariate dependence in clustering, where the UNL provides an interpretable measure of whether latent group structure can be explained by specific covariates. The methodology is illustrated on two real-world datasets.

stat.ME↗

arXiv2026

One Tool Is Enough: Reinforcement Learning for Repository-Level LLM Agents

Locating files and functions requiring modification in large software repositories is challenging due to their scale and structural complexity. Existing LLM-based methods typically treat this as a repository-level retrieval task and rely on multiple auxiliary tools, which often overlook code execution logic and complicate model control. We propose RepoNavigator, an LLM agent equipped with a single execution-aware tool: jumping to the definition of an invoked symbol. This unified design reflects the actual flow of code execution while simplifying tool manipulation. RepoNavigator is trained end-to-end via Reinforcement Learning (RL) directly from a base pretrained model, without relying on closed-source distillation. Experiments demonstrate that RL-trained RepoNavigator achieves state-of-the-art performance, with the 7B model outperforming 14B baselines, the 14B model surpassing 32B competitors, and the 32B model exceeding closed-source models such as GPT-5 on most metrics. These results confirm that integrating a single, structurally grounded tool with RL training provides an efficient and scalable solution for repository-level issue localization.

cs.SE↗

arXiv2026

RepoZero: Can LLMs Generate a Code Repository from Scratch?

Large Language Models (LLMs) have recently shown remarkable progress in code generation, yet their ability to construct complete software repositories from scratch remains poorly understood. A fundamental bottleneck is the lack of verifiable and scalable evaluation: existing benchmarks either focus on patch-based editing or rely on human or LLM-based judgments, which introduce bias and limit reproducibility. In this work, we present RepoZero, the first benchmark that enables fully automated, execution-based verification of repository-level generation from scratch. Our key idea is to reformulate generation as repository reproduction: given only API specifications, an agent must re-implement an entire repository such that its behavior matches the original implementation. This design allows for strict black-box validation via output equivalence, while naturally supporting large-scale construction by reusing existing open-source repositories. To further mitigate data leakage and shortcut solutions, we introduce cross-language constraints and a sandboxed evaluation protocol. Building on this benchmark, we propose an Agentic Code-Test Evolution (ACE) framework that performs iterative test generation and error-driven refinement, enabling effective test-time scaling for repository-level synthesis. Extensive experiments across multiple state-of-the-art LLMs and agent frameworks reveal that even the strongest LLM agents achieve only limited pass rates (30\% - 55\%), exposing a substantial gap between current capabilities and real-world software development requirements. Our results establish RepoZero as a challenging, scalable, and reliable testbed for end-to-end code generation, and highlight self-verification via test generation as a critical direction for advancing LLM-based coding agents.

cs.SE↗

arXiv2026

StormWave: An Open-Source Portable SDR Platform for Over-the-Air Resilience Evaluation of Terrestrial and Aerial Communications

This paper presents \emph{StormWave}, an open-source, portable software-defined Radio Frequency (RF) interference generation and monitoring platform designed for realistic field-based evaluation of the resilience of wireless communication systems. StormWave enables seamless composition and runtime switching among a wide range of narrowband and wideband waveforms, while supporting multiple digital modulations, adaptive coding, and multi-radio orchestration with real-time spectrum visualization. We evaluate the effectiveness of StormWave through both outdoor ground and air-to-air (A2A) experiments. Ground experiments demonstrate clear waveform- and modulation-dependent interference effects under realistic propagation conditions, while A2A experiments reveal pronounced distance-dependent constellation distortion and access-symbol degradation under active interference. The StormWave source code will be released to the community, with the expectation that StormWave will be used as a flexible, extensible, and field-ready platform for systematically validating interference resilience of wireless systems under realistic operating conditions.

eess.SY↗

arXiv2026

Assessing the Feasibility of a Video-Based Conversational Chatbot Survey for Measuring Perceived Cycling Safety: A Pilot Study in New York City

Bicycle safety is important for bikeability and transportation efficiency. However, conventional surveys often fall short in capturing how people actually perceive cycling environments because they rely heavily on respondents' recall rather than in-the-moment experience. By leveraging large language models (LLMs), this study proposes a new method of combining video-based surveys with a conversational AI chatbot to collect human perceptions of cycling safety and the reasons behind these perceptions. The paper developed the AI chatbot using a modular LLM architecture, integrating prompt engineering, state management, and rule-based control to support the structure of human-AI interaction. This paper evaluates the feasibility of the proposed video-based conversational chatbot using complete responses from sixteen participants to the pilot survey across nine street segments in New York City. The method feasibility was assessed using a seven-point scale rating for user experience (i.e., ease of use, supportiveness, efficiency) and a five-point scale for chatbot usability (i.e., personality, roboticness, friendliness), yielding positive results with mean scores of 5.00 out of 7 (standard deviation = 1.6) and 3.47 out of 5 (standard deviation = 0.43), respectively. The data feasibility was assessed using multiple techniques: (1) Natural language processing (NLP), such as KeyBERT, for overall safety and feature analysis to extract built-environment attributes; (2) K-means clustering for semantic analysis to identify reasons and suggestions; and (3) regression to estimate the effects of built-environment and demographic variables on perceived safety outcomes. The results show the potential of AI chatbots as a novel approach to collecting data on human perception, behavior, and future visions for transport planning.

cs.CY↗

arXiv2026

Exploring Sidewalk Sheds in New York City through Chatbot Surveys and Human Computer Interaction

Sidewalk sheds are a common feature of the streetscape in New York City, reflecting ongoing construction and maintenance activities. However, policymakers and local business owners have raised concerns about reduced storefront visibility and altered pedestrian navigation. Although sidewalk sheds are widely used for safety, their effects on pedestrian visibility and movement are not directly measured in current planning practices. To address this, we developed an AI-based chatbot survey that collects image-based annotations and route choices from pedestrians, linking these responses to specific shed design features, including clearance height, post spacing, and color. This AI chatbot survey integrates a large language model (e.g., Google's Gemini-1.5-flash-001 model) with an image-annotation interface, allowing users to interact with street images, mark visual elements, and provide structured feedback through guided dialogue. To explore pedestrian perceptions and behaviors, this paper conducts a grid-based analysis of entrance annotations and applies logistic mixed-effects modeling to assess sidewalk choice patterns. Analysis of the dataset (n = 25) shows that: (1) the presence of scaffolding significantly reduces pedestrians' ability to identify ground-floor retail entrances, and (2) variations in weather conditions and shed design features significantly influence sidewalk selection behavior. By integrating generative AI into urban research, this study demonstrates a novel method for evaluating sidewalk shed designs and provides empirical evidence to support adjustments to shed guidelines that improve the pedestrian experience without compromising safety.

cs.HC↗

arXiv2026

SciEGQA: A Dataset for Scientific Evidence-Grounded Question Answering and Reasoning

Scientific documents contain complex multimodal structures, which makes evidence localization and scientific reasoning in Document Visual Question Answering particularly challenging. However, most existing benchmarks evaluate models only at the page level without explicitly annotating the evidence regions that support the answer, which limits both interpretability and the reliability of evaluation. To address this limitation, we introduce SciEGQA, a scientific document question answering and reasoning dataset with semantic evidence grounding, where supporting evidence is represented as semantically coherent document regions annotated with bounding boxes. SciEGQA consists of two components: a **human-annotated fine-grained benchmark** containing 1,623 high-quality question--answer pairs, and a **large-scale automatically constructed training set** with over 30K QA pairs generated through an automated data construction pipeline. Extensive experiments on a wide range of Vision-Language Models (VLMs) show that existing models still struggle with evidence localization and evidence-based question answering in scientific documents. Training on the proposed dataset significantly improves the scientific reasoning capabilities of VLMs. The project page is available at https://yuwenhan07.github.io/SciEGQA-project/.

cs.DB↗

arXiv2026

CoDesignAI: An AI-Enabled Multi-Agent, Multi-User System for Collaborative Urban Design at the Conceptual Stage

Public participation has become increasingly important in collaborative urban design; yet, existing processes often face challenges in achieving efficient and scalable citizen engagement. To address this gap, this study explores how large language models (LLMs) can support cooperation among community members in participatory design. We introduce CoDesignAI, a collaborative urban design tool that combines multiple users, representing residents or stakeholders, with multiple AI agents, representing domain experts who provide facilitation and professional knowledge during the conceptual stage of urban design. This paper presents the system architecture and main components of the tool, illustrating how users interact with AI agents within a collaborative and iterative design workflow. Specifically, the system integrates generative AI with spatial mapping services to support street-level visualization of design proposals. AI agents assist users by summarizing discussion content, extracting shared design intentions, and generating prompts for presenting design interventions. The system also enables users to revise and refine their ideas over multiple rounds while documenting the design process. By combining conversational AI, multi-user interaction, and image-based design grounded in real-world urban contexts, this study argues that AI-enabled design systems can help shift urban design from an expert-centered practice to a more open and participatory process. The paper contributes a new web-based platform for AI-assisted collaborative design and offers an early exploration of how AI agents may expand the capacity for public participation in urban design.

cs.HC↗

arXiv2026

BenchLink: An SoC-Based Benchmark for Resilient Communication Links in GPS-Denied Environments

Accurate timing and synchronization, typically enabled by GPS, are essential for modern wireless communication systems. However, many emerging applications must operate in GPS-denied environments where signals are unreliable or disrupted, resulting in oscillator drift and carrier frequency impairments. To address these challenges, we present BenchLink, a System-on-Chip (SoC)-based benchmark for resilient communication links that functions without GPS and supports adaptive pilot density and modulation. Unlike traditional General Purpose Processor (GPP)-based software-defined radios (e.g. USRPs), the SoC-based design allows for more precise latency control. We implement and evaluate BenchLink on Zynq UltraScale+ MPSoCs, and demonstrate its effectiveness in both ground and aerial environments. A comprehensive dataset has also been collected under various conditions. We will make both the SoC-based link design and dataset available to the wireless community. BenchLink is expected to facilitate future research on data-driven link adaptation, resilient synchronization in GPS-denied scenarios, and emerging applications that require precise latency control, such as integrated radar sensing and communication.

eess.SP↗

arXiv2026

Bridging the Urban Divide: Adaptive Cross-City Learning for Disaster Sentiment Understanding

Social media platforms provide a real-time lens into public sentiment during natural disasters; however, models built solely on textual data often reinforce urban-centric biases and overlook underrepresented communities. This paper introduces an adaptive cross-city learning framework that enhances disaster sentiment understanding by integrating mobility-informed behavioral signals and city similarity-based data augmentation. Focusing on the January 2025 Southern California wildfires, our model achieves state-of-the-art performance and reveals geographically diverse sentiment patterns, particularly in areas experiencing overlapping fire exposure or delayed emergency responses. We further identify positive correlations between emotional expressions and real-world mobility shifts, underscoring the value of combining behavioral and textual features. Through extensive experiments, we demonstrate that multimodal fusion and city-aware training significantly improve both accuracy and fairness. Collectively, these findings highlight the importance of context-sensitive sentiment modeling and provide actionable insights toward developing more inclusive and equitable disaster response systems.

cs.SI↗

arXiv2026

Less Is More -- Until It Breaks: Security Pitfalls of Vision Token Compression in Large Vision-Language Models

Visual token compression is widely adopted to improve the inference efficiency of Large Vision-Language Models (LVLMs), enabling their deployment in latency-sensitive and resource-constrained scenarios. However, existing work has mainly focused on efficiency and performance, while the security implications of visual token compression remain largely unexplored. In this work, we first reveal that visual token compression substantially degrades the robustness of LVLMs: models that are robust under uncompressed inference become highly vulnerable once compression is enabled. These vulnerabilities are state-specific; failure modes emerge only in the compressed setting and completely disappear when compression is disabled, making them particularly hidden and difficult to diagnose. By analyzing the key stages of the compression process, we identify instability in token importance ranking as the primary cause of this robustness degradation. Small and imperceptible perturbations can significantly alter token rankings, leading the compression mechanism to mistakenly discard task-critical information and ultimately causing model failure. Motivated by this observation, we propose a Compression-Aware Attack to systematically study and exploit this vulnerability. CAA directly targets the token selection mechanism and induces failures exclusively under compressed inference. We further extend this approach to more realistic black-box settings and introduce Transfer CAA, where neither the target model nor the compression configuration is accessible. We further evaluate potential defenses and find that they provide only limited protection. Extensive experiments across models, datasets, and compression methods show that visual token compression significantly undermines robustness, revealing a previously overlooked efficiency-security trade-off.

cs.CR↗

arXiv2025

Training Report of TeleChat3-MoE

TeleChat3-MoE is the latest series of TeleChat large language models, featuring a Mixture-of-Experts (MoE) architecture with parameter counts ranging from 105 billion to over one trillion,trained end-to-end on Ascend NPU cluster. This technical report mainly presents the underlying training infrastructure that enables reliable and efficient scaling to frontier model sizes. We detail systematic methodologies for operator-level and end-to-end numerical accuracy verification, ensuring consistency across hardware platforms and distributed parallelism strategies. Furthermore, we introduce a suite of performance optimizations, including interleaved pipeline scheduling, attention-aware data scheduling for long-sequence training,hierarchical and overlapped communication for expert parallelism, and DVM-based operator fusion. A systematic parallelization framework, leveraging analytical estimation and integer linear programming, is also proposed to optimize multi-dimensional parallelism configurations. Additionally, we present methodological approaches to cluster-level optimizations, addressing host- and device-bound bottlenecks during large-scale training tasks. These infrastructure advancements yield significant throughput improvements and near-linear scaling on clusters comprising thousands of devices, providing a robust foundation for large-scale language model development on hardware ecosystems.

cs.CL↗

arXiv2025

Population-Evolve: a Parallel Sampling and Evolutionary Method for LLM Math Reasoning

Test-time scaling has emerged as a promising direction for enhancing the reasoning capabilities of Large Language Models in last few years. In this work, we propose Population-Evolve, a training-free method inspired by Genetic Algorithms to optimize LLM reasoning. Our approach maintains a dynamic population of candidate solutions for each problem via parallel reasoning. By incorporating an evolve prompt, the LLM self-evolves its population in all iterations. Upon convergence, the final answer is derived via majority voting. Furthermore, we establish a unification framework that interprets existing test-time scaling strategies through the lens of genetic algorithms. Empirical results demonstrate that Population-Evolve achieves superior accuracy with low performance variance and computational efficiency. Our findings highlight the potential of evolutionary strategies to unlock the reasoning power of LLMs during inference.

cs.AI↗

arXiv2025

Character-Level Perturbations Disrupt LLM Watermarks

Large Language Model (LLM) watermarking embeds detectable signals into generated text for copyright protection, misuse prevention, and content detection. While prior studies evaluate robustness using watermark removal attacks, these methods are often suboptimal, creating the misconception that effective removal requires large perturbations or powerful adversaries. To bridge the gap, we first formalize the system model for LLM watermark, and characterize two realistic threat models constrained on limited access to the watermark detector. We then analyze how different types of perturbation vary in their attack range, i.e., the number of tokens they can affect with a single edit. We observe that character-level perturbations (e.g., typos, swaps, deletions, homoglyphs) can influence multiple tokens simultaneously by disrupting the tokenization process. We demonstrate that character-level perturbations are significantly more effective for watermark removal under the most restrictive threat model. We further propose guided removal attacks based on the Genetic Algorithm (GA) that uses a reference detector for optimization. Under a practical threat model with limited black-box queries to the watermark detector, our method demonstrates strong removal performance. Experiments confirm the superiority of character-level perturbations and the effectiveness of the GA in removing watermarks under realistic constraints. Additionally, we argue there is an adversarial dilemma when considering potential defenses: any fixed defense can be bypassed by a suitable perturbation strategy. Motivated by this principle, we propose an adaptive compound character-level attack. Experimental results show that this approach can effectively defeat the defenses. Our findings highlight significant vulnerabilities in existing LLM watermark schemes and underline the urgency for the development of new robust mechanisms.

cs.CR↗

arXiv2025

Technical Report of TeleChat2, TeleChat2.5 and T1

We introduce the latest series of TeleChat models: \textbf{TeleChat2}, \textbf{TeleChat2.5}, and \textbf{T1}, offering a significant upgrade over their predecessor, TeleChat. Despite minimal changes to the model architecture, the new series achieves substantial performance gains through enhanced training strategies in both pre-training and post-training stages. The series begins with \textbf{TeleChat2}, which undergoes pretraining on 10 trillion high-quality and diverse tokens. This is followed by Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) to further enhance its capabilities. \textbf{TeleChat2.5} and \textbf{T1} expand the pipeline by incorporating a continual pretraining phase with domain-specific datasets, combined with reinforcement learning (RL) to improve performance in code generation and mathematical reasoning tasks. The \textbf{T1} variant is designed for complex reasoning, supporting long Chain-of-Thought (CoT) reasoning and demonstrating substantial improvements in mathematics and coding. In contrast, \textbf{TeleChat2.5} prioritizes speed, delivering rapid inference. Both flagship models of \textbf{T1} and \textbf{TeleChat2.5} are dense Transformer-based architectures with 115B parameters, showcasing significant advancements in reasoning and general task performance compared to the original TeleChat. Notably, \textbf{T1-115B} outperform proprietary models such as OpenAI's o1-mini and GPT-4o. We publicly release \textbf{TeleChat2}, \textbf{TeleChat2.5} and \textbf{T1}, including post-trained versions with 35B and 115B parameters, to empower developers and researchers with state-of-the-art language models tailored for diverse applications.

cs.CL↗

arXiv2025

No Free Lunch: Rethinking Internal Feedback for LLM Reasoning

Reinforcement learning has emerged as a powerful paradigm for post-training large language models (LLMs) to improve reasoning. Approaches like Reinforcement Learning from Human Feedback (RLHF) and Reinforcement Learning with Verifiable Rewards (RLVR) have shown strong results, but they require extensive external supervision. We investigate an alternative class of methods, Reinforcement Learning from Internal Feedback (RLIF), which relies solely on intrinsic model-derived signals instead of external rewards. In particular, we leverage unsupervised reward proxies such as token-level entropy, trajectory-level entropy, and self-certainty. Our theoretical analysis shows these internal objectives are partially equivalent, and we empirically evaluate various RLIF strategies on challenging math reasoning benchmarks. Experimental results demonstrate that RLIF can boost the reasoning performance of base LLMs at the beginning phase of the training, matching or surpassing RLVR techniques on these tasks. However, when training progresses, performance degrades even below the model before training. Moreover, we find that RLIF yields little improvement for instruction-tuned models, indicating diminishing returns of intrinsic feedback once an LLM is already instruction-tuned. We further analyze this limitation by mixing model weights and explain the reason of RLIF's training behaviors, providing practical guidelines for integrating internal feedback signals into LLM training. We hope our analysis of internal feedback will inform more principled and effective strategies for LLM post-training.

cs.LG↗