Distilling Image Prototypes for Guided Test-Time Adaptation
为解决测试时适应中的错误累积和源知识遗忘问题，本文提出DIPTTA框架，通过动态生成特征原型和校准不确定性估计来提高模型鲁棒性。
0 citationsRead paper
Selected work
Representative Papers
Visual graphs for image classification: does the structure affect performance?
Existing deep learning models struggle to effectively encode spatial, topological, and semantic structural information inherent in images. This work systematically evaluates the impact of various visual graph construction strategies on image classification performance within a unified three-layer Graph Convolutional Network (GCN) framework. For the first time, it demonstrates that the graph structure itself plays a decisive role in model performance. The study underscores the critical importance of the graph construction preprocessing stage, providing empirical evidence that well-designed graph structures substantially enhance classification accuracy. These findings offer both methodological guidance and practical justification for graph structure selection and preprocessing in visual graph neural networks.
0 citationsRead paper
Application of association rule mining to assess forest species distribution in Italy considering abiotic and biotic factors
Addressing the need for dynamic assessment of forest biodiversity and ecological conservation in Italy, this study tackles the challenge of uncovering latent ecological relationships from complex, multi-source environmental and vegetation data. Method: We propose the first analytical framework applying Association Rule Mining (ARM) to forest ecosystems, leveraging data from 6,784 plots—including plant community composition, geospatial information, bioclimatic indices, soil properties, and remote-sensing variables—and employing the FP-Growth algorithm to extract species–environment and interspecific co-occurrence rules. Contribution/Results: The approach reveals interpretable, data-driven ecological associations, identifying keystone “hub” species and their strong environmental responses—for instance, *Picea abies* exhibits high-confidence associations with temperature and precipitation seasonality (confidence: 90.9%; lift: 7.13). These findings advance mechanistic understanding of species coexistence, inform evidence-based territorial planning, and enhance ecosystem resilience under global change.
0 citationsRead paper
Recent publications
Latest Papers
Distilling Image Prototypes for Guided Test-Time Adaptation
为解决测试时适应中的错误累积和源知识遗忘问题，本文提出DIPTTA框架，通过动态生成特征原型和校准不确定性估计来提高模型鲁棒性。
0 citationsRead paper
Visual graphs for image classification: does the structure affect performance?
Existing deep learning models struggle to effectively encode spatial, topological, and semantic structural information inherent in images. This work systematically evaluates the impact of various visual graph construction strategies on image classification performance within a unified three-layer Graph Convolutional Network (GCN) framework. For the first time, it demonstrates that the graph structure itself plays a decisive role in model performance. The study underscores the critical importance of the graph construction preprocessing stage, providing empirical evidence that well-designed graph structures substantially enhance classification accuracy. These findings offer both methodological guidance and practical justification for graph structure selection and preprocessing in visual graph neural networks.
0 citationsRead paper
Application of association rule mining to assess forest species distribution in Italy considering abiotic and biotic factors
Addressing the need for dynamic assessment of forest biodiversity and ecological conservation in Italy, this study tackles the challenge of uncovering latent ecological relationships from complex, multi-source environmental and vegetation data. Method: We propose the first analytical framework applying Association Rule Mining (ARM) to forest ecosystems, leveraging data from 6,784 plots—including plant community composition, geospatial information, bioclimatic indices, soil properties, and remote-sensing variables—and employing the FP-Growth algorithm to extract species–environment and interspecific co-occurrence rules. Contribution/Results: The approach reveals interpretable, data-driven ecological associations, identifying keystone “hub” species and their strong environmental responses—for instance, *Picea abies* exhibits high-confidence associations with temperature and precipitation seasonality (confidence: 90.9%; lift: 7.13). These findings advance mechanistic understanding of species coexistence, inform evidence-based territorial planning, and enhance ecosystem resilience under global change.
0 citationsRead paper