Institution profile

Southwest Jiao Tong University

Academic institutionasia · cn
Official website
Research library337linked papers
Opportunities0open roles
Selected work

Representative Papers

A 3D virtual geographic environment for flood representation towards risk communication

Apr 01, 2024International Journal of Applied Earth Observation and Geoinformation

Existing flood risk research overly relies on specialized numerical models, hindering comprehension and application by non-expert stakeholders and thus limiting the effectiveness of risk communication. To address this, this paper proposes a three-dimensional visualization framework integrating Virtual Geographic Environments (VGE) with flood risk communication. The framework synthesizes hydrological simulation outputs, multi-source geospatial data, 3D modeling, GIS, and virtual reality technologies to enable spatiotemporally dynamic, interactive, and photorealistic flood inundation scenario visualization. It overcomes the limitations of conventional static two-dimensional representations, substantially enhancing the intuitiveness, accessibility, and public cognitive efficiency of flood risk information. As a result, it provides an operational, user-centered visualization support tool for public engagement and emergency decision-making—constituting a significant methodological innovation in transitioning flood risk communication from an expert-driven to a user-centered paradigm.

19 citationsRead paper

Streaming Hallucination Detection in Long Chain-of-Thought Reasoning

Jan 05, 2026arXiv.org

This work proposes the first streaming hallucination detection framework for long-chain-of-thought (CoT) reasoning, addressing the challenge that hallucinations in such settings are often subtle, propagate across reasoning steps, and are difficult to detect and localize in real time. The approach models hallucination as a dynamically evolving latent state throughout the reasoning process, leveraging step-level judgments as local observations to construct prefix-accumulated signals that trace the global evolution of hallucinatory behavior. By doing so, the method enables real-time, interpretable monitoring of hallucinations in extended CoT sequences and provides fine-grained evidential support for detected anomalies. This significantly enhances both the timeliness and transparency of hallucination detection, offering a principled and practical solution for improving the reliability of complex reasoning systems.

1 citationsRead paper

From Failure to Mastery: Generating Hard Samples for Tool-use Agents

Jan 04, 2026arXiv.org

Existing agent training data predominantly consist of simple, homogeneous trajectories that fail to capture complex implicit logical dependencies, thereby limiting model performance. This work proposes HardGen, an automated agent pipeline that leverages failure cases to generate high-quality synthetic data. HardGen is the first framework to transform failed trajectories into dynamic, structured API graphs, integrating modular high-level tools with conditional priors to produce verifiable, complex reasoning chains (Chain-of-Thought). A closed-loop evaluation feedback mechanism continuously refines sample quality. A 4B-parameter model trained on HardGen-synthesized data outperforms leading open- and closed-source large language models—including GPT-5.2, Gemini-3-Pro, and Claude-Opus-4.5—across multiple benchmarks, demonstrating significant improvements in complex tool usage and reasoning capabilities.

1 citationsRead paper

GenCAMO: Scene-Graph Contextual Decoupling for Environment-aware and Mask-free Camouflage Image-Dense Annotation Generation

Jan 03, 2026arXiv.org

This work addresses the scarcity of high-quality, large-scale annotated camouflage imagery that hinders dense prediction tasks such as camouflaged object detection and open-vocabulary segmentation. To overcome this limitation, we propose an environment-aware, mask-free generative framework that leverages scene graph context disentanglement to jointly synthesize realistic multimodal camouflaged images along with their dense annotations—including depth maps, attribute descriptions, and textual prompts—thereby constructing GenCAMO-DB, the first large-scale synthetic dataset for this domain. Experimental results demonstrate that models trained on our synthesized data achieve significantly improved performance in complex camouflaged scenarios, validating both the effectiveness and generalization capability of the generated data.

1 citationsRead paper

A Simple and Efficient Approach to Batch Bayesian Optimization

Nov 25, 2024

Existing batch Bayesian optimization (BO) methods suffer significant performance degradation as batch size increases, failing to fully exploit parallel computing resources. To address this scalability bottleneck, we propose a novel paradigm for large-scale parallel BO: it decomposes the high-dimensional search space into orthogonal, axis-aligned low-dimensional subspaces and introduces the first-of-its-kind Expected Subspace Improvement (ESI) acquisition function, which jointly optimizes diverse yet convergent batch query points within each subspace. This approach effectively balances exploration and exploitation while enabling scalable parallelization. Empirical evaluation on standard benchmarks demonstrates that our method substantially outperforms sequential BO in wall-clock time and consistently achieves state-of-the-art or competitive performance among seven leading batch BO algorithms. The implementation is publicly available in MATLAB, confirming both efficiency and practical applicability.

1 citationsRead paper
Recent publications

Latest Papers

Opacity Is Not Just Opacity

Sep 13, 2026

该研究通过扩展alpha合成系数域解决颜色对比度问题，提出了一种新的适应背景的对比度增强方法。

0 citationsRead paper