Research map / Graph learning / GNN architectures

Bias, Measurement Error, and Double-Dipping: When Can GNN Convolutions Help Brain Connectome Prediction?

NeurIPS 2026 · Graph learning › GNN architectures · approach: Simple and scalable GNNs

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.

Analyzes why GNN graph convolutions often fail to beat graph-agnostic baselines for fMRI-based brain connectome prediction, framing it statistically around bias, measurement error, and double-dipping.

Problem

GNNs for fMRI-based prediction, where empirical functional connectivity matrices serve as both node features and graph structure, often fail to outperform simple graph-agnostic baselines.

Approach

Proposes a statistical explanation for the failure by formulating connectome prediction as a statistical estimation problem, attributing it to bias, measurement error in the empirical connectivity, and double-dipping (reusing the same connectivity data for graph construction and features); the paper characterizes when GNN convolutions can help.

Surprising findings

• Recent GNN studies and the authors' own experiments show GNNs often fail to outperform simpler graph-agnostic baselines on connectome prediction.

Experimental setup

Baselines: graph-agnostic baselines

Closest papers by meaning

• NeuroGraph: Benchmarks for Graph Machine Learning in Brain Connectomics (NeurIPS 2023 · similarity 0.91)
• MAGNET: Manifold-Aware Graph Diffusion Network for Connectome Generation (NeurIPS 2026 · similarity 0.87)
• Long-range Brain Graph Transformer (NeurIPS 2024 · similarity 0.87)
• Neurodegenerative Brain Network Classification via Adaptive Diffusion with Temporal Regularization (ICML 2024 · similarity 0.87)
• Feature Distribution on Graph Topology Mediates the Effect of Graph Convolution: Homophily Perspective (ICML 2024 · similarity 0.86)
• On the Stability of Graph Convolutional Neural Networks: A Probabilistic Perspective (NeurIPS 2025 · similarity 0.86)