Special Sessions

Machine learning techniques combined with Earth observation data are transforming the way we monitor and address global sustainability challenges. This track focuses on recent advances in machine learning methods and their applications in remote sensing, environmental monitoring, climate analysis, natural resource management, disaster assessment, and sustainable urban and agricultural development. Topics of interest include deep learning for satellite imagery, geospatial data analytics, environmental prediction models, AI-driven decision support systems, and interdisciplinary approaches for achieving sustainability goals using Earth observation technologies.

With the continuous advancement of deep learning technologies, efficient data representation and cross-domain modeling capabilities have become core drivers in the field of intelligent perception. This special session focuses on the frontier topics of pose estimation and human behavior recognition, aiming to bring together researchers working in deep learning, multimodal fusion, and computer vision. We place particular emphasis on the theories, methods, and applications of pattern recognition for human skeletal structures, motion trajectories, and interactive behaviors in complex scenarios. The session welcomes contributions on spatiotemporal manifold modeling, cross-view pose alignment, multisensor fusion, and behavior representation learning on non-Euclidean spaces such as graph structures. Furthermore, we are also interested in explorations of foundation models and large-scale model techniques for human behavior understanding, including action semantic mining, few-shot behavior classification, and the prediction of behavioral intentions for pedestrians and drivers in embodied intelligence and autonomous driving applications.

With the rapid development of artificial intelligence, computer vision, pattern recognition, and data science, an increasing amount of data exhibits complex geometric structures that cannot be adequately characterized by conventional vector representations in Euclidean space. Examples include Symmetric Positive Definite (SPD) matrices, Correlation matrices, subspaces, graph-structured data, and hierarchical semantic representations. In this context, non-Euclidean representation learning has emerged as an important research direction in machine learning and artificial intelligence.

Non-Euclidean representation learning aims to overcome the limitations of conventional Euclidean representations by developing data representations, feature learning, metric learning, attention mechanisms, and deep neural networks directly on spaces with non-Euclidean geometric structures. Such geometric representations can provide a more faithful characterization of the local and global structures inherent in complex data. Recent advances in Riemannian manifold learning, hyperbolic representation learning, spherical representation learning, and mixed-curvature representation learning have demonstrated significant potential in a wide range of applications, including computer vision, brain signal analysis, medical imaging, natural language processing, graph learning, and multimodal learning.

This special track aims to bring together researchers working on non-Euclidean representation learning and related areas to exchange recent advances in theoretical foundations, geometric modeling, representation learning, deep learning architectures, optimization, and real-world applications. Particular emphasis will be placed on the integration of non-Euclidean geometry with modern artificial intelligence, including novel representation learning theories, geometry-aware deep learning architectures, optimization and learning algorithms on non-Euclidean spaces, as well as emerging applications in various scientific and engineering domains.

We invite researchers to submit original research contributions addressing new theories, methodologies, algorithms, and applications related to non-Euclidean representation learning. We hope this special track will provide a platform for exploring emerging challenges and opportunities at the intersection of geometry, representation learning, and artificial intelligence.