Innovative Geometry-Aware GNN Solutions

We specialize in advanced geometric neural networks, focusing on hyperbolic and spherical models for improved data representation and analysis.

A large geodesic sphere with a complex triangular panel pattern against a clear blue sky, with lush green palm leaves in the corner.

A dark, futuristic sphere surrounded by a geometric wireframe and adorned with shiny metallic spheres. The structure suggests a blend of technology and art, featuring intricate patterns and reflective surfaces.

A geometric, paper-crafted object with a complex, polyhedral shape is suspended against a beige background. The object appears to be made from pieces of paper with map-like details, featuring lines and small symbols.

Our Research Phases

Our approach includes theoretical modeling, algorithm development, and experimental validation, ensuring robust performance in complex data tasks like node classification.

Geometric GNNs

Innovative models for non-Euclidean graph neural networks development.

A close-up view of a geodesic sphere structure with a triangular panel pattern illuminated with blue, purple, and yellow lights against a dark night sky.

Hyperbolic GNN

Adaptive curvature learning in hyperbolic space structures.

A three-dimensional sphere with a hexagonal pattern is placed against a background of similar hexagonal tiles. The perspective creates a sense of depth and texture, enhanced by the subtle use of light and shadows.

Spherical GNN

Modeling cyclic dependencies using spherical space closure.

Geometric metal spheres are interconnected by large cylindrical beams against a cloudy sky background. The mirrored surfaces of the spheres reflect light and obscure shapes, creating a futuristic architectural design.

Various geometric shapes such as spheres, cylinders, and ellipses are floating against a light blue background. The objects feature a gradient of colors including blue, pink, and gold, giving them a metallic and reflective appearance.

Experimental Validation

Multi-task experiments on public datasets for performance evaluation.

Algorithm Development

Designing models for hierarchical and cyclic structures.

A 3D rendering of a large, smooth, metallic cube floating in a soft, light-colored space. Spherical objects are attached to the cube, creating a minimalist and futuristic appearance.

Geometric Neural Networks

We specialize in advanced non-Euclidean graph neural networks, focusing on hyperbolic and spherical geometries to enhance machine learning performance across various applications.