Nonlocal operator learning for fMRI encoding and decoding tasks

2026-05-22 · Source: cs.LG updates on arXiv.org · Field: Science & Research — Health & Medical Research, Mathematics & Computational Sciences · Depth: Expert, extended

Summary

A nonlocal neural integral operator framework has been developed for fMRI encoding and decoding tasks, utilizing a latent neural integral operator that performs fixed point iterations. Evaluated on two open-source fMRI datasets, Haxby [HGF+01] and Miyawaki [MUY+08], the model demonstrates that larger temporal windows consistently improve performance and yield more structured learned representations. In decoding experiments, the learned latent space often provided clearer class separation compared to raw data. For encoding tasks, despite moderate absolute performance (R^2 ≈ 0.25, Pearson correlation ≈ 0.46), longer temporal windows still produced consistent gains. Additionally, whole-brain recordings significantly increased decoding performance on the Haxby dataset, achieving the highest scores among tested models with 20 time points. The framework also enhanced self-supervised classification of stimulus structure, particularly for shorter temporal windows.

Key takeaway

For Research Scientists or AI Scientists developing fMRI analysis models, you should consider integrating nonlocal neural integral operators. This approach significantly improves encoding and decoding performance, particularly when leveraging broader spatiotemporal context like longer temporal windows and whole-brain recordings. Your models will likely achieve clearer latent-space organization and stronger class separability, especially for challenging tasks or limited temporal information.

Key insights

Nonlocal neural integral operators effectively model fMRI dynamics, benefiting from broader spatiotemporal context for improved prediction and representation.

Principles

• Brain dynamics exhibit nonlocal spatiotemporal dependencies.
• Broader temporal context improves fMRI modeling.
• Model architectures should capture distributed dependencies.

Method

The framework implements a latent neural integral operator performing fixed point iterations in an auxiliary space, followed by a decoder for classification and stimuli prediction.

In practice

• Use longer temporal windows for fMRI analysis.
• Incorporate whole-brain data for decoding tasks.
• Employ nonlocal operators for distributed brain dynamics.

Topics

• fMRI Encoding
• fMRI Decoding
• Neural Integral Operators
• Nonlocal Operators
• Spatiotemporal Context
• Brain Dynamics Modeling

Code references

• nilearn/nilearn

Best for: AI Scientist, Research Scientist

Related on AIssential

• Multiple Neural Operators Achieve Near-Optimal Rates for Multi-Task Learning — Shared representations in multi-task operator learning achieve near-optimal rates without increasing overall cost.
• Spectral Audit of In-Context Operator Networks — Prediction accuracy alone is insufficient for neural operators; local operator fidelity requires spectral auditing.
• Harness In-Context Operator Learning with Chain of Operators — Harnessing frozen neural operators with explicit transformations improves out-of-distribution generalization without fine-tuning.
• Modeling Local, Global, and Cross-Modal Context in Multimodal 3D MRI — Explicitly modeling intra- and cross-modal interactions is crucial for maximizing the utility of multimodal 3D brain MRI.
• Learning Physical Operators using Neural Operators — A physics-informed framework improves neural operator generalization and temporal continuity for PDE solving via operator splitting.
• How Much Memory Do We Need? Adaptive Memory Gate for Neural Operators — AMGFNO dynamically adjusts memory weights in neural operators, improving performance across varying observation conditions.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.LG updates on arXiv.org.