MENDR: Manifold Explainable Neural Data Representations

2025-08-06

AI Summary

Current EEG foundation models suffer from opaque pretraining procedures and lack interpretable, reconstructable embeddings, hindering clinical deployment. To address this, we propose FilterRiemann—a novel architecture integrating filter-bank design with a Riemannian manifold Transformer. Methodologically, we pioneer the coupling of discrete wavelet packet transform with symmetric positive definite (SPD) matrix embedding to explicitly model multi-scale EEG features on the Riemannian manifold; EEG representations are geometrically encoded as visually interpretable ellipsoidal structures, enabling precise signal reconstruction. This design uniquely combines non-Euclidean modeling capability with deterministic, traceable inference. Evaluated across multiple clinical EEG tasks, FilterRiemann achieves near-state-of-the-art performance with significantly reduced parameter count, while delivering high transparency, strong interpretability, and end-to-end reconstructability.

Technology Category

• Machine Learning: Learning with Manifolds
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Computer Vision: Multi-modal Vision

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization

Abstract

Foundation models for electroencephalography (EEG) signals have recently demonstrated success in learning generalized representations of EEGs, outperforming specialized models in various downstream tasks. However, many of these models lack transparency in their pretraining dynamics and offer limited insight into how well EEG information is preserved within their embeddings. For successful clinical integration, EEG foundation models must ensure transparency in pretraining, downstream fine-tuning, and the interpretability of learned representations. Current approaches primarily operate in the temporal domain, overlooking advancements in digital signal processing that enable the extraction of deterministic and traceable features, such as wavelet-based representations. We propose MENDR (Manifold Explainable Neural Data Representations), a filter bank-based EEG foundation model built on a novel Riemannian Manifold Transformer architecture to resolve these issues. MENDR learns symmetric positive definite matrix embeddings of EEG signals and is pretrained on a large corpus comprising over 4,000 hours of EEG data, decomposed via discrete wavelet packet transforms into multi-resolution coefficients. MENDR significantly enhances interpretability by visualizing symmetric positive definite embeddings as geometric ellipsoids and supports accurate reconstruction of EEG signals from learned embeddings. Evaluations across multiple clinical EEG tasks demonstrate that MENDR achieves near state-of-the-art performance with substantially fewer parameters, underscoring its potential for efficient, interpretable, and clinically applicable EEG analysis.

Problem

• Lack of transparency in EEG foundation models' pretraining dynamics
• Limited interpretability of learned EEG signal embeddings
• Overlooking deterministic features like wavelet-based representations

Innovation

• Riemannian Manifold Transformer architecture
• Discrete wavelet packet transforms decomposition
• Symmetric positive definite matrix embeddings