Research map / Graph learning / GNN architectures

Matrix-Free Stochastic Training of Low-Rank Spectral Graph Learning via Randomized Adaptive Spectral Estimation

NeurIPS 2026 · Graph learning › GNN architectures · approach: Spectral GNN design

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes a matrix-free, stochastically trainable low-rank spectral graph learning method using randomized adaptive spectral estimation, avoiding full-graph eigendecomposition.

Problem

Explicit spectral layers for large-graph learning are hard to train stochastically because their spectral bases require a full-graph spectralization step and their graph-wide contexts are full-batch; matrix-free polynomial filters train stochastically but tie response to a finite polynomial degree.

Approach

Learns low-rank spectral graph filters in a matrix-free way by estimating spectral components randomly and adaptively during stochastic training, so no full-graph eigensolver or full-batch spectral context is needed.

Experimental setup

Builds on: polynomial spectral graph filters

Closest papers by meaning

• HoloNets: Spectral Convolutions do extend to Directed Graphs (ICLR 2024 · similarity 0.89)
• Graph Few-Shot Learning via Adaptive Spectrum Experts and Cross-Set Distribution Calibration (NeurIPS 2025 · similarity 0.89)
• Position: Spectral GNNs Rely Less on Graph Fourier Basis than Conceived (ICML 2025 · similarity 0.88)
• Unifying Homophily and Heterophily for Spectral Graph Neural Networks via Triple Filter Ensembles (NeurIPS 2024 · similarity 0.88)
• A Unified Uncertainty Representation for Graph Neural Networks via Doubly-Spectral Stochastic Expansion (NeurIPS 2026 · similarity 0.88)
• Minimax Rates and Spectral Distillation for Tree Ensembles (NeurIPS 2026 · similarity 0.88)