RECTOR: Masked Region-Channel-Temporal Modeling for Affective and Cognitive Representation Learning

2026-06-13 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Health & Medical Research, Mental Health & Psychological Support · Depth: Expert, quick

Summary

RECTOR (Masked Region-Channel-Temporal Modeling) is an end-to-end self-supervised framework designed for robust representation learning from EEG/sEEG data, specifically addressing affective and cognitive disorders. It unifies joint region-channel-temporal representation learning, moving beyond fixed anatomical priors. At its core, RECTOR-SA employs hierarchical, block-sparse self-attention, which is induced by Adaptive Functional Partitioning to evolve region structures into adaptive functional regions. The self-supervision mechanism, called Masked Topology and Representation Learning, optimizes three complementary objectives: Masked Predictive Modeling, Topological Structure Modeling, and Cross-View Consistency. This framework achieves new state-of-the-art results in EEG emotion recognition and sEEG task-engagement classification across diverse benchmarks. Its strong robustness to missing channels and cross-montage generalization highlights its potential for large-scale pre-training on heterogeneous EEG/sEEG data, offering interpretable insights at both region and channel levels.

Key takeaway

For AI Scientists and Research Scientists developing diagnostic tools for affective and cognitive disorders, RECTOR offers a robust self-supervised framework for EEG/sEEG data. You should consider integrating its adaptive functional partitioning and multi-objective masked learning approach to improve model generalization and interpretability, especially when dealing with heterogeneous datasets or missing channels. This could significantly enhance the accuracy and reliability of your brain network dynamics analysis.

Key insights

RECTOR unifies region-channel-temporal EEG/sEEG representation learning via self-supervision, adapting functional regions for robust affective and cognitive disorder diagnosis.

Principles

• Adaptive functional partitioning improves EEG/sEEG region structures.
• Self-supervision enhances robustness to missing data.
• Joint optimization of predictive, topological, and cross-view objectives.

Method

RECTOR-SA uses hierarchical, block-sparse self-attention with Adaptive Functional Partitioning. Masked Topology and Representation Learning optimizes Masked Predictive Modeling, Topological Structure Modeling, and Cross-View Consistency.

In practice

• Apply RECTOR for EEG emotion recognition.
• Use RECTOR for sEEG task-engagement classification.
• Pre-train on heterogeneous EEG/sEEG datasets.

Topics

• EEG/sEEG Representation Learning
• Self-Supervised Learning
• Affective Computing
• Cognitive Disorders
• Brain Network Dynamics
• Masked Modeling
• Adaptive Functional Partitioning

Best for: AI Scientist, Research Scientist

Related on AIssential

• Continual Learning for fMRI-Based Brain Disorder Diagnosis via Functional Connectivity Matrices Generative Replay — A continual learning framework uses generative replay and knowledge distillation to diagnose brain disorders from sequential fMRI data.
• Neural Response to Familiar Names Predicts Outcome of Comatose ICU Patients: A Prospective Observational Cohort Study — EEG responses to familiar names predict outcomes in comatose ICU patients.
• Modeling of ASD/TD Children's Behaviors in Interaction with a Virtual Social Robot During a Music Education Program Using Deep Neural Networks — Deep neural networks can effectively model and distinguish ASD and neurotypical child behaviors during robot interaction.
• Fusion Learning from Dynamic Functional Connectivity: Combining the Amplitude and Phase of fMRI Signals to Identify Brain Disorders — Combining fMRI signal amplitude and phase via fusion learning improves brain disorder detection.
• GSEM: Graph-based Self-Evolving Memory for Experience Augmented Clinical Reasoning — GSEM organizes clinical experiences into a dual-layer memory graph for improved decision-making and retrieval.
• Transition-Based Digital Twin Modelling for Alzheimer's Disease under Sparse Longitudinal Data — Local transition modeling offers a data-efficient and robust strategy for personalized disease forecasting in neurodegenerative disorders.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.