YHYu Huang

Yu Huang

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Yu Huang has a total of 202 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Beyond One Granularity: Spectral Ensemble and Consensus Expansion for Training-Free Composed Image Retrieval

    by Dexuan Xu, Yifan Xiao, Shijie Li, Xingyu Yang, Siqi Hou, Hanpin Wang, Yu Huang

    Composed image retrieval (CIR) enables expressive multimodal search: given a reference image and a text modifier, the goal is to retrieve images that match the visual content as altered by the text. Building effective CIR systems has historically required expensive triplet-annotated training data, limiting scalability. Training-free CIR sidesteps this bottleneck by repurposing pre-trained vision-language models to compose queries without any task-specific supervision. However, existing training-free methods commit to a single, globally fixed projection granularity when aligning visual and textual features. This creates an inherent trade-off: coarse granularities capture broad categorical distinctions but lose fine-grained detail, while fine granularities recover instance-level attributes but amplify noise. The problem is especially acute in instance-level CIR, where per-query databases vary from fewer than 10 to over 10,000 images, making any fixed setting suboptimal across the evaluation. We therefore propose a unified retrieval method with two complementary components that fuse scores across multiple granularities to capture complementary evidence and yield more robust rankings. MACE (Multi-granularity Adaptive Composition with Expansion) constructs a spectral ensemble from multiple nested truncation levels and fuses their scores without any additional eigendecomposition or database modification. CGCQE (Cross-Granularity Consensus Query Expansion) integrates into MACE to strengthen pseudo-relevance feedback, selecting expansion candidates that rank consistently well across all granularities and preventing granularity-specific false positives from corrupting the expanded query. Experiments on five CIR benchmarks, including the challenging instance-level i-CIR, show that MACE with CGCQE achieves state-of-the-art performance, outperforming both training-free and supervised zero-shot methods.

  • Open Access

    AI for Science: A Systematic Survey of Architectures, Autonomy, and Open Challenges

    by Yongqiang Zhao, Yuhan Liu, Yuqian Han, Huanyao Zhang, Yanfang Zhou, Ziliang Wang, Yu Huang, Yan Zhang, Yonghong Tian

    Artificial Intelligence for Science (AI4S) is a pivotal research frontier revolutionizing scientific discovery, yet it suffers from fragmentation and lacks system-level exploration. This survey unifies its landscape: it defines AI4S against related concepts, verifies its status as the fifth scientific paradigm, and presents a six-layer framework and five-tier autonomy taxonomy. It also summarizes key techniques, core challenges and future directions, aiming to guide relevant research and infrastructure development.

  • Open Access

    Fuzzy-Gated Verification of LLM Agents for Industrial Fault Diagnosis: An Evolutionary-Optimized Closed-Loop Framework

    by Yifan Xiao, Dexuan Xu, Shijie Li, Weiping Ding, Yu Huang

    Intelligent fault diagnosis and machine health management are essential for reliable operation of industrial systems, yet deploying large language models (LLMs) for these tasks introduces reliability risks including hallucinated evidence, overconfident outputs, and inconsistent reasoning. We propose CL-LLMOps, a closed-loop LLMOps framework that orchestrates LLM agents through five integrated stages—from alarm normalization to regression-tested action. At the semantic verification level—where, to our knowledge, soft computing has not previously been applied—its hierarchical mechanism combines rule-based pattern checking, cross-attention evidence scoring, and Mamdani fuzzy inference gating, jointly optimized via multi-objective particle swarm optimization, to detect hallucinations and evidence contradictions before maintenance decisions. Evaluation on three public datasets spanning power quality classification, bearing fault diagnosis, and wind turbine SCADA anomaly detection—validated across two open-source LLM backbones (Llama-3-8B and Mistral-7B)—shows that CL-LLMOps achieves 98.6% diagnostic accuracy on power quality data, 99.3% on CWRU bearing data, and 95.2% on CARE wind turbine data. The framework reduces inconsistency-type hallucination from 12% to 2% relative to the fine-tuned LLM baseline while maintaining 93% action accuracy and 96% risk compliance on PQ data, with fuzzy gating specifically improving action accuracy by 5 percentage points through graded borderline handling. This work bridges LLM-powered intelligence and operational dependability for industrial deployment, offering a novel integration of established soft computing techniques—neural attention, fuzzy logic, and evolutionary optimization—at the semantic verification level for deploying trustworthy AI agents in fault diagnosis under the Industry 5.0 paradigm.

  • Open Access

    Probabilistic anticipation of AMOC transitions under critical forcing magnitudes and rates via deep learning

    by Wenjie Zhang, Yu Huang, Sebastian Bathiany, Yechul Shin, Maya Ben-Yami, Suiping Zhou, Niklas Boers

  • Open Access

    Comment on egusphere-2026-1376

    by Ming Shi, Huang Yu, Zuntao Fu, Hao Xue, Xiaolong Qiu, Xintong Hao, Xiaohan Zhang, Haiyang Yu, Weiqi Xing

    Sea surface temperature anomaly (SSTA) fields over the tropical Pacific exhibit complex spatiotemporal variability that traditional regional average indices may fail to capture. In this study, we take the evolution of daily SSTA fields as a dynamical system (DS) and compute the DS instantaneous dimension (DSID) to characterize local dynamical properties. Our results reveal a significant shift in the mean value (SMV) of DSID around the year 2007 over the western tropical Pacific (notably the Niño 4 and Niño 3.4 regions), a phenomenon that is absent in the eastern Pacific.To explain this region-dependent SMV of DSID in SSTA fields, we examine the sensitivity of DSID to the changes of the system states using idealized system models, such low-dimensional (Lorenz-63) and high-dimensional (Lorenz-96) systems. Results demonstrate that the estimation of DSID is sensitive to variations in both high-frequency variability and system states,for both low-dimensional and high-dimensional systems. These findings from idealized systems provide a theoretical basis for interpreting the observed SMV in SSTA fields. Further analysis reveals that the SMV in DSID can be largely attributed to changes in high-frequency variability across different time spans. Collectively, these results advance our understanding of the physical interpretation of DSID and offer valuable insights for the study of complex real-world systems.

  • Open Access

    Comment on egusphere-2026-1376

    by Ming Shi, Huang Yu, Zuntao Fu, Hao Xue, Xiaolong Qiu, Xintong Hao, Xiaohan Zhang, Haiyang Yu, Weiqi Xing

    Sea surface temperature anomaly (SSTA) fields over the tropical Pacific exhibit complex spatiotemporal variability that traditional regional average indices may fail to capture. In this study, we take the evolution of daily SSTA fields as a dynamical system (DS) and compute the DS instantaneous dimension (DSID) to characterize local dynamical properties. Our results reveal a significant shift in the mean value (SMV) of DSID around the year 2007 over the western tropical Pacific (notably the Niño 4 and Niño 3.4 regions), a phenomenon that is absent in the eastern Pacific.To explain this region-dependent SMV of DSID in SSTA fields, we examine the sensitivity of DSID to the changes of the system states using idealized system models, such low-dimensional (Lorenz-63) and high-dimensional (Lorenz-96) systems. Results demonstrate that the estimation of DSID is sensitive to variations in both high-frequency variability and system states,for both low-dimensional and high-dimensional systems. These findings from idealized systems provide a theoretical basis for interpreting the observed SMV in SSTA fields. Further analysis reveals that the SMV in DSID can be largely attributed to changes in high-frequency variability across different time spans. Collectively, these results advance our understanding of the physical interpretation of DSID and offer valuable insights for the study of complex real-world systems.