Implicit Graph Neural Network
Implicit Graph Neural Networks (IGNNs) aim to improve upon traditional GNNs by efficiently capturing long-range dependencies within graph data through iterative processes converging to a fixed-point representation. Current research focuses on developing more efficient training algorithms, addressing issues like over-smoothing and limited effective range, and exploring architectures that incorporate multiscale information or handle dynamic graphs. These advancements enhance the performance and applicability of IGNNs across various tasks, including node and graph classification, and solving partial differential equations, particularly in scenarios with complex graph structures and boundary conditions.
Papers
October 22, 2024
October 11, 2024
June 25, 2024
August 7, 2023
February 6, 2023
November 19, 2022