I\textsuperscript{2}RiMA: Spectral Riemannian Representation with Temporal Attention for Mental Stress Detection based on EEG Signals

2026-07-01 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Neuroscience & Biomedical AI · Depth: Expert, quick

Summary

I²RiMA, an Intra-Inter Riemannian Manifold Attention Network, is a novel method for EEG-based mental stress detection designed to overcome challenges posed by subject-dependent and frequency-specific stress patterns. It addresses limitations of conventional Riemannian methods by constructing spatial covariance matrices independently at each frequency point and mapping them to the SPD tangent space, which preserves channel-wise geometry and frequency-specific cues. The network also incorporates frequency cluster aggregation to select informative spectral components and reduce redundancy, forming data-driven frequency clusters aligned with EEG rhythms. Furthermore, an intra-inter slice attention module adaptively integrates local spectral dynamics and global temporal context across EEG sequences. Experiments across three datasets demonstrate that I²RiMA consistently outperforms five state-of-the-art baselines, achieving up to 82.78% balanced accuracy while maintaining efficiency with only 1.60M parameters and 31.95M FLOPs.

Key takeaway

For Machine Learning Engineers developing EEG-based mental stress detection systems, I²RiMA presents a highly effective and efficient architecture. Its novel approach of combining spectral Riemannian representation with temporal attention addresses critical challenges in subject-dependent and frequency-specific patterns. You should consider integrating its frequency-specific covariance matrix construction and intra-inter slice attention module to enhance model accuracy and robustness in real-world applications, especially when working with limited computational resources given its 1.60M parameters and 31.95M FLOPs.

Key insights

I²RiMA uses spectral Riemannian representation and temporal attention for robust, efficient EEG-based mental stress detection.

Principles

• Discriminative stress patterns are subject-dependent and frequency-specific.
• Neural oscillations are critical for high-level cognitive state decoding.
• Preserving channel-wise geometry and frequency-specific cues improves detection.

Method

I²RiMA constructs frequency-specific spatial covariance matrices, maps them to SPD tangent space, aggregates frequency clusters, and applies intra-inter slice attention for EEG stress detection.

In practice

• Apply frequency-specific covariance for EEG signal analysis.
• Utilize attention mechanisms to integrate temporal context.
• Reduce redundancy via data-driven frequency clustering.

Topics

• EEG Signal Processing
• Mental Stress Detection
• Riemannian Geometry
• Attention Networks
• Spectral Representation
• Covariance Matrices

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• A spectral audit framework reveals task-dependent aperiodic reliance across EEG and ECG deep learning — Deep learning models on physiological data often rely on aperiodic signals, impacting task performance.
• Gating Enables Curvature: A Geometric Expressivity Gap in Attention — Multiplicative gating in attention enables non-flat geometric representations, enhancing expressivity for nonlinear tasks.
• Bayesian Spectral Emotion Transition Discovery from Multi-Annotator Disagreement — Preserving annotator uncertainty in multi-rater judgments is crucial for understanding emotion dynamics and aligning with psychological theory.
• KAN-MLP-Mixer: A comprehensive investigation of the usage of Kolmogorov-Arnold Networks (KANs) for improving IMU-based Human Activity Recognition — Strategically combining KANs for data embedding and classification with MLPs for feature mixing significantly enhances noisy IMU-based HAR.
• Boundary-aware Prototype-driven Adversarial Alignment for Cross-Corpus EEG Emotion Recognition — The PAA framework improves cross-corpus EEG emotion recognition by aligning class-conditional distributions and refining decision boundaries.
• Asymptotic Signal Subspace Recovery in Softmax Attention Models — Softmax attention mechanisms can rigorously extract hidden signals from high-dimensional noise through gradient-based learning dynamics.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.