arXiv subjects

Shu Xu

Publications and source records attributed to Shu Xu.

16 recordsLinked to original sources

arXiv2026

Impacts of Nuclear War on Human Health from Changed Surface Ultraviolet Radiation

Climate model simulations indicate that surface ultraviolet (UV) radiation would change after soot injection into the stratosphere during a nuclear war, due to the competing effects of ozone depletion and aerosol attenuation. Using climate model simulations, we evaluate UV impacts under two scenarios: a regional India-Pakistan conflict producing 5 Tg of soot and a global U.S.-Russia war producing 150 Tg. UV enhancements due to ozone depletion substantially shorten safe outdoor exposure time, particularly for individuals with lighter skin types. Applying UV dose-response relationships to the year 2000 population data, without accounting for direct conflict mortality or famine-related population loss, the 5 Tg scenarios result in approximately 5,300-9,800 additional skin cancer deaths within 10-15 years and up to 75,000 cumulative excess deaths over the following century. In contrast, strong aerosol attenuation under the 150 Tg scenario initially suppresses surface UV, resulting in about 8,500 fewer skin cancer deaths within 15 years and a maximum cumulative reduction of approximately 17,000 deaths over the following century. These findings demonstrate that nuclear war-induced changes in surface UV radiation represent a persistent but previously understudied health impact of nuclear war. While skin cancer would not dominate overall mortality following a nuclear war, enhanced cumulative surface UV radiation represents an additional, long-lasting threat to human health that compounds other global impacts such as climate disruption and food insecurity. The excess UV may also pose negative impacts on animals and plants, including those used for agriculture, which remain to be quantified.

physics.ao-ph↗

arXiv2026

KAT-Coder-V2.5 Technical Report

We present KAT-Coder-V2.5, a coding-focused agentic model trained to act autonomously inside real, executable repositories rather than as a single-turn code generator. Its capability is bottlenecked less by model scale than by the scarcity of reproducible environments, verifiable rewards, and high-value trajectories, which we address with an end-to-end agentic post-training framework. AutoBuilder reconstructs multilingual repositories into sandboxed environments with fail-to-pass and pass-to-pass verification at scale, from which we regenerate self-contained task specifications, recover near-miss trajectories, and distill supervision through process-aware filtering, while KwaiClawEnv synthesizes large-scale tool-use trajectories from executable services and real task seeds. We further scale reinforcement learning with harness randomization, a reliability-hardened sandbox, an asymmetric actor--critic PPO with hindsight-augmented value estimation, and a harness-oriented reward framework, and unify SWE, Agent-Claw, and WebCoding experts via Multi-Teacher On-Policy Distillation. Across six software-engineering and agentic benchmarks, KAT-Coder-V2.5 delivers the best agentic tool-use result on PinchBench and ranks second only to the frontier Opus 4.8 on repository-level software engineering. Our service is available at https://streamlake.com/product/kat-coder.

cs.SE↗

arXiv2026

A Self-Evolving Framework for Efficient Terminal Agents via Observational Context Compression

As terminal agents scale to long-horizon, multi-turn workflows, a key bottleneck is not merely limited context length, but the accumulation of noisy terminal observations in the interaction history. Retaining raw observations preserves useful environment feedback, but also leads to context saturation and high token cost; conversely, naive compression may discard task-critical signals needed for subsequent actions. Because terminal environments are highly heterogeneous across repositories, commands, and execution states, heuristic-based or fixed-prompt compression methods are difficult to generalize. We propose TACO, a plug-and-play, training-free, self-evolving Terminal Agent Compression framework for existing terminal agents. TACO automatically discovers, refines, and reuses structured compression rules from interaction trajectories, enabling workflow-adaptive filtering of low-value terminal outputs while preserving task-relevant observations. Experiments on TerminalBench (TB 1.0 and TB 2.0) and four additional terminal-related benchmarks, including SWE-Bench Lite, CompileBench, DevEval, and CRUST-Bench, show that TACO consistently improves task performance and token efficiency across agent scaffolds and backbone models. On TerminalBench, TACO yields 1%-4% accuracy gains across strong agentic models and improves accuracy by around 2%-3% under the same token budget. On additional terminal-related benchmarks, it reduces total token consumption while maintaining or improving task success rates. These results suggest that self-evolving, workflow-adaptive observation compression is an effective path toward more reliable and efficient long-horizon terminal agents. The code is publicly available at https://github.com/multimodal-art-projection/TACO.

cs.CL↗

arXiv2026

Story2Proposal: A Scaffold for Structured Scientific Paper Writing

Generating scientific manuscripts requires maintaining alignment between narrative reasoning, experimental evidence, and visual artifacts across the document lifecycle. Existing language-model generation pipelines rely on unconstrained text synthesis with validation applied only after generation, often producing structural drift, missing figures or tables, and cross-section inconsistencies. We introduce Story2Proposal, a contract-governed multi-agent framework that converts a research story into a structured manuscript through coordinated agents operating under a persistent shared visual contract. The system organizes architect, writer, refiner, and renderer agents around a contract state that tracks section structure and registered visual elements, while evaluation agents supply feedback in a generate evaluate adapt loop that updates the contract during generation. Experiments on tasks derived from the Jericho research corpus show that Story2Proposal achieved an expert evaluation score of 6.145 versus 3.963 for DirectChat (+2.182) across GPT, Claude, Gemini, and Qwen backbones. Compared with the structured generation baseline Fars, Story2Proposal obtained an average score of 5.705 versus 5.197, indicating improved structural consistency and visual alignment.

cs.CL↗

arXiv2026

JSSAnet: Theory-Guided Subchannel Partitioning and Joint Spatial Attention for Near-Field Channel Estimation

The deployment of extremely large-scale antenna array (ELAA) in sixth-generation (6G) communication systems introduces unique challenges for efficient near-field channel estimation. To tackle these issues, this paper presents a theory-guided approach that incorporates angular information into an attention-based estimation framework. A piecewise Fourier representation is proposed to implicitly encode the near-field channel's inherent nonlinearity, enabling the entire channel to be segmented into multiple subchannels, each mapped to the angular domain via the discrete Fourier transform (DFT). Then, we develop a joint subchannel-spatial-attention network (JSSAnet) to extract the spatial features of both intra- and inter-subchannels. To guide theoretically the design of the joint attention mechanism, we derive upper and lower bounds based on approximation criteria and DFT quantization loss mitigation, respectively. Following by both bounds, a JSSA layer of an attention block is constructed to assign independent and adaptive spatial attention weights to each subchannel in parallel. Subsequently, a feed-forward network (FFN) of an attention block further captures and refines the residual nonlinear dependencies across subchannels. Moreover, the proposed JSSA map is linearly computed via element-wise product combining large-kernel convolutions (DLKC), maintaining strong contextual learning capability. Numerical results verify the effectiveness of embedding sparsity information into the attention network and demonstrate JSSAnet achieves superior estimation performance compared with existing methods.

eess.SP↗

arXiv2026

Search More, Think Less: Rethinking Long-Horizon Agentic Search for Efficiency and Generalization

Recent deep research agents primarily improve performance by scaling reasoning depth, but this leads to high inference cost and latency in search-intensive scenarios. Moreover, generalization across heterogeneous research settings remains challenging. In this work, we propose \emph{Search More, Think Less} (SMTL), a framework for long-horizon agentic search that targets both efficiency and generalization. SMTL replaces sequential reasoning with parallel evidence acquisition, enabling efficient context management under constrained context budgets. To support generalization across task types, we further introduce a unified data synthesis pipeline that constructs search tasks spanning both deterministic question answering and open-ended research scenarios with task appropriate evaluation metrics. We train an end-to-end agent using supervised fine-tuning and reinforcement learning, achieving strong and often state of the art performance across benchmarks including BrowseComp (48.6\%), GAIA (75.7\%), Xbench (82.0\%), and DeepResearch Bench (45.9\%). Compared to Mirothinker-v1.0, SMTL with maximum 100 interaction steps reduces the average number of reasoning steps on BrowseComp by 70.7\%, while improving accuracy.

cs.CL↗

arXiv2026

Idea2Story: An Automated Pipeline for Transforming Research Concepts into Complete Scientific Narratives

Autonomous scientific discovery with large language model (LLM)-based agents has recently made substantial progress, demonstrating the ability to automate end-to-end research workflows. However, existing systems largely rely on runtime-centric execution paradigms, repeatedly reading, summarizing, and reasoning over large volumes of scientific literature online. This on-the-spot computation strategy incurs high computational cost, suffers from context window limitations, and often leads to brittle reasoning and hallucination. We propose Idea2Story, a pre-computation-driven framework for autonomous scientific discovery that shifts literature understanding from online reasoning to offline knowledge construction. Idea2Story continuously collects peer-reviewed papers together with their review feedback, extracts core methodological units, composes reusable research patterns, and organizes them into a structured methodological knowledge graph. At runtime, underspecified user research intents are aligned to established research paradigms, enabling efficient retrieval and reuse of high-quality research patterns instead of open-ended generation and trial-and-error. By grounding research planning and execution in a pre-built knowledge graph, Idea2Story alleviates the context window bottleneck of LLMs and substantially reduces repeated runtime reasoning over literature. We conduct qualitative analyses and preliminary empirical studies demonstrating that Idea2Story can generate coherent, methodologically grounded, and novel research patterns, and can produce several high-quality research demonstrations in an end-to-end setting. These results suggest that offline knowledge construction provides a practical and scalable foundation for reliable autonomous scientific discovery.

cs.CE↗

arXiv2025

Parabolic hysteresis problems revisited: Finite element error analysis and convergent Newton-type solvers

Numerical investigations of partial differential equations with hysteresis have largely focused on simulations, leaving numerical error analysis unexplored and relying mainly on derivative-free nonlinear solvers. This work establishes rigorous finite element error estimates for the backward Euler fully discrete scheme applied to semilinear and quasilinear parabolic equations involving continuous hysteresis operators. To efficiently handle the inherent nonsmoothness of the resulting nonlinear algebraic systems, we develop a damped smoothing Newton solver under a general condition on the smoothing approximation, ensuring global convergence together with local Q-quadratic convergence. Numerical experiments confirm the theoretical convergence rates for semilinear problems, while showing higher-than-predicted orders for quasilinear ones. The robustness and efficiency of the proposed solver are further demonstrated in comparison with existing methods.

math.NA↗

arXiv2025

A Multi-Scale Spatial Attention Network for Near-field MIMO Channel Estimation

The deployment of extremely large-scale array (ELAA) brings higher spectral efficiency and spatial degree of freedom, but triggers issues on near-field channel estimation. Existing near-field channel estimation schemes primarily exploit sparsity in the transform domain. However, these schemes are sensitive to the transform matrix selection and the stopping criteria. Inspired by the success of deep learning (DL) in far-field channel estimation, this paper proposes a novel spatial-attention-based method for reconstructing extremely large-scale MIMO (XL-MIMO) channel. Initially, the spatial antenna correlations of near-field channels are analyzed as an expectation over the angle-distance space, which demonstrate correlation range of an antenna element varies with its position. Due to the strong correlation between adjacent antenna elements, interactions of inter-subchannel are applied to describe inherent correlation of near-field channels instead of inter-element. Subsequently, a multi-scale spatial attention network (MsSAN) with the inter-subchannel correlation learning capabilities is proposed tailed to near-field MIMO channel estimation. We employ the multi-scale architecture to refine the subchannel size in MsSAN. Specially, we inventively introduce the sum of dot products as spatial attention (SA) instead of cross-covariance to weight subchannel features at different scales in the SA module. Simulation results are presented to validate the proposed MsSAN achieves remarkable the inter-subchannel correlation learning capabilities and outperforms others in terms of near-field channel reconstruction.

eess.SP↗

arXiv2025

Efficient Beam Selection for ISAC in Cell-Free Massive MIMO via Digital Twin-Assisted Deep Reinforcement Learning

Beamforming enhances signal strength and quality by focusing energy in specific directions. This capability is particularly crucial in cell-free integrated sensing and communication (ISAC) systems, where multiple distributed access points (APs) collaborate to provide both communication and sensing services. In this work, we first derive the distribution of joint target detection probabilities across multiple receiving APs under false alarm rate constraints, and then formulate the beam selection procedure as a Markov decision process (MDP). We establish a deep reinforcement learning (DRL) framework, in which reward shaping and sinusoidal embedding are introduced to facilitate agent learning. To eliminate the high costs and associated risks of real-time agent-environment interactions, we further propose a novel digital twin (DT)-assisted offline DRL approach. Different from traditional online DRL, a conditional generative adversarial network (cGAN)-based DT module, operating as a replica of the real world, is meticulously designed to generate virtual state-action transition pairs and enrich data diversity, enabling offline adjustment of the agent's policy. Additionally, we address the out-of-distribution issue by incorporating an extra penalty term into the loss function design. The convergency of agent-DT interaction and the upper bound of the Q-error function are theoretically derived. Numerical results demonstrate the remarkable performance of our proposed approach, which significantly reduces online interaction overhead while maintaining effective beam selection across diverse conditions including strict false alarm control, low signal-to-noise ratios, and high target velocities.

cs.ET↗

arXiv2024

Optimal-rate error estimates and a twice decoupled solver for a backward Euler finite element scheme of the Doyle-Fuller-Newman model of lithium-ion cells

We investigate the convergence of a backward Euler finite element discretization applied to a multi-domain and multi-scale elliptic-parabolic system, derived from the Doyle-Fuller-Newman model for lithium-ion cells. We establish optimal-order error estimates for the solution in the norms $l^2(H^1)$ and $l^2(L^2(H^q_r))$, $q=0,1$. To improve computational efficiency, we propose a novel solver that accelerates the solution process and controls memory usage. Numerical experiments with realistic battery parameters validate the theoretical error rates and demonstrate the significantly superior performance of the proposed solver over existing solvers.

math.NA↗

arXiv2024

Optimal convergence in finite element semi-discrete error analysis of the Doyle-Fuller-Newman model beyond 1D with a novel projection operator

We present a finite element semi-discrete error analysis for the Doyle-Fuller-Newman model, which is the most popular model for lithium-ion batteries. Central to our approach is a novel projection operator designed for the pseudo-($N$+1)-dimensional equation, offering a powerful tool for multiscale equation analysis. Our results bridge a gap in the analysis for dimensions $2 \le N \le 3$ and achieve optimal convergence rates of $h+(\Delta r)^2$. Additionally, we perform a detailed numerical verification, marking the first such validation in this context. By avoiding the change of variables, our error analysis can also be extended beyond isothermal conditions.

math.NA↗

arXiv2024

MindSpore Quantum: A User-Friendly, High-Performance, and AI-Compatible Quantum Computing Framework

We introduce MindSpore Quantum, a pioneering hybrid quantum-classical framework with a primary focus on the design and implementation of noisy intermediate-scale quantum (NISQ) algorithms. Leveraging the robust support of MindSpore, an advanced open-source deep learning training/inference framework, MindSpore Quantum exhibits exceptional efficiency in the design and training of variational quantum algorithms on both CPU and GPU platforms, delivering remarkable performance. Furthermore, this framework places a strong emphasis on enhancing the operational efficiency of quantum algorithms when executed on real quantum hardware. This encompasses the development of algorithms for quantum circuit compilation and qubit mapping, crucial components for achieving optimal performance on quantum processors. In addition to the core framework, we introduce QuPack, a meticulously crafted quantum computing acceleration engine. QuPack significantly accelerates the simulation speed of MindSpore Quantum, particularly in variational quantum eigensolver (VQE), quantum approximate optimization algorithm (QAOA), and tensor network simulations, providing astonishing speed. This combination of cutting-edge technologies empowers researchers and practitioners to explore the frontiers of quantum computing with unprecedented efficiency and performance.

quant-ph↗

arXiv2022

Artificial Intelligence Security Competition (AISC)

The security of artificial intelligence (AI) is an important research area towards safe, reliable, and trustworthy AI systems. To accelerate the research on AI security, the Artificial Intelligence Security Competition (AISC) was organized by the Zhongguancun Laboratory, China Industrial Control Systems Cyber Emergency Response Team, Institute for Artificial Intelligence, Tsinghua University, and RealAI as part of the Zhongguancun International Frontier Technology Innovation Competition (https://www.zgc-aisc.com/en). The competition consists of three tracks, including Deepfake Security Competition, Autonomous Driving Security Competition, and Face Recognition Security Competition. This report will introduce the competition rules of these three tracks and the solutions of top-ranking teams in each track.

cs.CR↗

arXiv2022

Nonlinear quantum gates for a Bose-Einstein condensate

Quantum interferometry and quantum information processing have been proposed for Bose-Einstein condensates (BECs), but BECs are described in complicated ways such as using quantum field theory or using a nonlinear differential equation. Nonlinear quantum mechanics does not mesh well with the superposition principle at the heart of interferometry and quantum information processing but could be compatible. Thus, we develop a rigorous foundation for quantum gates, obtained by solving the equation for evolution, and then we employ this foundation, combined with quantum-control techniques and appropriate state-sampling techniques, to devise feasible nonlinear Hadamard gates and thereby feasible, i.e., high-contrast, nonlinear Ramsey interferometry. Our approach to BEC interferometry and quantum logic shifts the paradigm by enlarging to the case of nonlinear quantum mechanics, which we apply to the cases of BEC interferometry and quantum information processing.

quant-ph↗

arXiv2018

Quantum walk on a chimera graph

We analyze a continuous-time quantum walk on a chimera graph, which is a graph of choice for designing quantum annealers, and we discover beautiful quantum-walk features such as localization that starkly distinguishes classical from quantum behavior. Motivated by technological thrusts, we study continuous-time quantum walks on enhanced variants of the chimera graph and on a diminished chimera graph with a random removal of sites. We explain the quantum walk by constructing a generating set for a suitable subgroup of graph isomorphisms and corresponding symmetry operators that commute with the quantum-walk Hamiltonian; the Hamiltonian and these symmetry operators provide a complete set of labels for the spectrum and the stationary states. Our quantum-walk characterization of the chimera graph and its variants yields valuable insights into graphs used for designing quantum-annealers.

quant-ph↗