Selected work

Representative Papers

CG-GLORE: A Conjugate Gradient-Based Global-Local Regularization Network for Sparse-View CT Reconstruction

Aug 15, 2026

This study addresses the ill-posed inverse problem in sparse-view CT reconstruction and the poor convergence of existing deep learning methods by proposing a compact deep unfolding framework inspired by second-order optimization. By constructing a structured Hessian proxy with a conjugate gradient solver and designing a global-local regularization module that integrates convolutional features with Nyström attention, the method effectively models image priors. Experiments on AAPM and DeepLesion datasets demonstrate stable convergence, significant noise power reduction, and enhanced visual fidelity. Achieving superior quantitative metrics compared to state-of-the-art approaches, this work provides an efficient and reliable solution for sparse-view CT reconstruction.

0 citationsRead paper

Coliseum project: Correlating climate change data with the behavior of heritage materials

Nov 17, 2025

Climate change is accelerating the weathering of cultural heritage materials; however, the nonlinear, multivariate coupling nature of degradation processes impedes identification and quantification of climatic drivers. To address this, we deployed microclimate sensor networks across three French heritage sites—including Strasbourg Cathedral—to synchronously collect meteorological, high-resolution imaging, chemical, and geospatial data. We propose a “weathering matrix” framework grounded in meteorological indices and establish a heterogeneous, multi-temporal data fusion architecture. Furthermore, we develop an AI-driven dynamic association model enabling quantitative prediction of material degradation trends under varying climate scenarios. Critically, this approach is the first to explicitly embed microclimatic response mechanisms into weathering modeling, thereby substantially enhancing predictive interpretability and spatiotemporal adaptability. The methodology advances heritage conservation from reactive intervention toward proactive, climate-resilient management.

0 citationsRead paper

Federated Learning for Video Violence Detection: Complementary Roles of Lightweight CNNs and Vision-Language Models for Energy-Efficient Use

Nov 10, 2025

To address privacy preservation, high energy consumption, and non-IID data challenges in federated video violence detection, this paper proposes a privacy–energy-efficiency co-optimization framework. It integrates a lightweight 3D CNN with a vision-language model (LLaVA-NeXT-Video-7B), marking the first comparative study of LoRA-finetuned VLMs versus personalized CNNs in federated settings. A hierarchical semantic classification mechanism enhances multi-class recognition, while a scene-complexity-aware dynamic model invocation strategy enables zero-shot inference and semantic-aware class grouping. Experiments demonstrate a binary classification accuracy exceeding 90%, a 58% reduction in 3D CNN energy consumption (from 240 Wh to 570 Wh), an 81% multi-class accuracy for the VLM, and a ROC AUC of 92.59%. The framework thus achieves a balanced trade-off among detection accuracy, system sustainability, and deployment flexibility.

0 citationsRead paper

Recent publications

Latest Papers

CG-GLORE: A Conjugate Gradient-Based Global-Local Regularization Network for Sparse-View CT Reconstruction

Aug 15, 2026

This study addresses the ill-posed inverse problem in sparse-view CT reconstruction and the poor convergence of existing deep learning methods by proposing a compact deep unfolding framework inspired by second-order optimization. By constructing a structured Hessian proxy with a conjugate gradient solver and designing a global-local regularization module that integrates convolutional features with Nyström attention, the method effectively models image priors. Experiments on AAPM and DeepLesion datasets demonstrate stable convergence, significant noise power reduction, and enhanced visual fidelity. Achieving superior quantitative metrics compared to state-of-the-art approaches, this work provides an efficient and reliable solution for sparse-view CT reconstruction.

0 citationsRead paper

Coliseum project: Correlating climate change data with the behavior of heritage materials

Nov 17, 2025

Climate change is accelerating the weathering of cultural heritage materials; however, the nonlinear, multivariate coupling nature of degradation processes impedes identification and quantification of climatic drivers. To address this, we deployed microclimate sensor networks across three French heritage sites—including Strasbourg Cathedral—to synchronously collect meteorological, high-resolution imaging, chemical, and geospatial data. We propose a “weathering matrix” framework grounded in meteorological indices and establish a heterogeneous, multi-temporal data fusion architecture. Furthermore, we develop an AI-driven dynamic association model enabling quantitative prediction of material degradation trends under varying climate scenarios. Critically, this approach is the first to explicitly embed microclimatic response mechanisms into weathering modeling, thereby substantially enhancing predictive interpretability and spatiotemporal adaptability. The methodology advances heritage conservation from reactive intervention toward proactive, climate-resilient management.

0 citationsRead paper

Federated Learning for Video Violence Detection: Complementary Roles of Lightweight CNNs and Vision-Language Models for Energy-Efficient Use

Nov 10, 2025

To address privacy preservation, high energy consumption, and non-IID data challenges in federated video violence detection, this paper proposes a privacy–energy-efficiency co-optimization framework. It integrates a lightweight 3D CNN with a vision-language model (LLaVA-NeXT-Video-7B), marking the first comparative study of LoRA-finetuned VLMs versus personalized CNNs in federated settings. A hierarchical semantic classification mechanism enhances multi-class recognition, while a scene-complexity-aware dynamic model invocation strategy enables zero-shot inference and semantic-aware class grouping. Experiments demonstrate a binary classification accuracy exceeding 90%, a 58% reduction in 3D CNN energy consumption (from 240 Wh to 570 Wh), an 81% multi-class accuracy for the VLM, and a ROC AUC of 92.59%. The framework thus achieves a balanced trade-off among detection accuracy, system sustainability, and deployment flexibility.

0 citationsRead paper