IGNN-Solver: A Graph Neural Solver for Implicit Graph Neural Networks

Name: IGNN-Solver: A Graph Neural Solver for Implicit Graph Neural Networks
Uploaded: 2026-10-20T20:25:50.728962+00:00
Duration: 663 s
Description: The IGNN-Solver is a novel graph neural solver designed to speed up fixed-point iterations in Implicit Graph Neural Networks (IGNNs). It uses a tiny Graph Neural Network to learn iterative updates via generalized Anderson Acceleration. This approach achieves significant inference speedups, ranging from 1.5x to 8x, on both small and large graphs without losing accuracy.

The IGNN-Solver is a novel graph neural solver designed to speed up fixed-point iterations in Implicit Graph Neural Networks (IGNNs). It uses a tiny Graph Neural Network to learn iterative updates via generalized Anderson Acceleration. This approach achieves significant inference speedups, ranging from 1.5x to 8x, on both small and large graphs without losing accuracy.

Open the full episode