JEPA · 2026-10-09
SPERA: Spherical Prior EEG Foundation Model with Geometry- and Frequency-Aware Latent Prediction
Minsu Kim, Ye-Sung Kim, Hyeseong Jeon, Wooseok Hyung, Joshua Lee, Chang-Hwan Im
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
SPERA is an EEG foundation model that uses a JEPA-style latent prediction objective to predict masked target latents from visible context. It adds a Legendre-polynomial spherical prior into attention to encode electrode geometry across varying montages, plus a factorized/hybrid attention backbone and a relational spectral regularizer. Pretrained on ~80,000 hours from 29,048 subjects across 106 datasets, the authors report the highest average balanced accuracy over nine downstream EEG tasks.
Why it matters
EEG models often struggle to generalize across heterogeneous subjects, devices, and electrode montages. SPERA’s latent-space JEPA objective, geometry-aware Legendre spherical prior in attention, and frequency-aware relational spectral regularizer are designed to address montage variability and capture spectral relationships without reconstructing artifacts. The paper evaluates across nine clinical/cognitive/BCI datasets and studies scaling and parameter efficiency under linear probing.
Method
- JEPA-style spatio-temporal latent prediction: student encodes context tokens; teacher (EMA) encodes full sequence; a cross-attention predictor maps context latents to masked target latents using an L1 latent loss.
- Geometry-aware attention: per-token coordinate embeddings plus a Legendre-polynomial spherical prior; truncated Legendre pairwise bias for factorized attention and anchor-based Legendre encoding for full-attention blocks.
- Frequency-aware latents: a relational spectral regularizer aligns pairwise similarity structure of predicted latents with pairwise similarity structure derived from log-power spectra (1–45 Hz).
Limitation
The paper does not state a limitation in the provided text.
Abstract (from arXiv)
Electroencephalography (EEG) provides a non-invasive measure of ongoing neural activity, but building general-purpose EEG models remains challenging due to the heterogeneity of subjects, devices, and electrode montages. Existing EEG foundation models predominantly rely on reconstruction-based objectives defined on the observed signal, which contains both neural and non-neural components. We introduce SPERA (Spherical Prior EEG Representation Architecture), an EEG foundation model that adopts the joint-embedding predictive architecture (JEPA) to predict in latent space. SPERA introduces a Legendre-polynomial spatial prior, incorporated into attention to encode varying scalp electrode geometries. Two further components adapt the model to EEG: factorized temporal and spatial attention interleaved with periodic full-attention blocks, and a relational spectral regularizer aligning latent similarity structure with spectral views. Pretrained on approximately 80,000 hours of EEG from 29,048 subjects across 106 datasets, SPERA achieves the highest average balanced accuracy across nine downstream tasks spanning clinical, cognitive, and BCI applications. SPERA further exhibits strong parameter efficiency under linear probing and robustness across varying recording conditions, suggesting its potential as a general-purpose backbone for diverse EEG analyses.
Related papers
- Controlled Evaluation of EEG Foundation Models: Random Initialisation, Band-Power and Validation-Selected Probes
- What masking geometry works best for EEG foundation models?
- Similar Predictive Fit but Different Latent Dynamics: Characterizing Learned Dynamical Structure in Personalized Models of Brain Disorders
- NeuroLens: Learning Latent Embeddings of Neural Semantics from Chronic Recordings
- Phenomenon-Graph JEPA: Label-Efficient Representation Learning for Contactless Cardiorespiratory Sensing
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-04 · Indexed by Pub-AI: 2026-10-09 UTC
SPERA: Spherical Prior EEG Foundation Model with Geometry- and Frequency-Aware Latent Prediction
Preprint posted: 2026-10-04 · Original record ↗
- arxiv
- 2610.10571
Explicit source versions (1)
- v1 · 2026-10-04
- Preprint available
Source date: 2026-10-04 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Electroencephalography (EEG) provides a non-invasive measure of ongoing neural activity, but building general-purpose EEG models remains challenging due to the heterogeneity of subjects, devices, and electrode montages. Existing EEG foundation models predominantly rely on reconstruction-based objectives defined on the observed signal, which contains both neural and non-neural components. We introduce SPERA (Spherical Prior EEG Representation Architecture), an EEG foundation model that adopts the joint-embedding predictive architecture (JEPA) to predict in latent space. SPERA introduces a Legendre-polynomial spatial prior, incorporated into attention to encode varying scalp electrode geometries. Two further components adapt the model to EEG: factorized temporal and spatial attention interleaved with periodic full-attention blocks, and a relational spectral regularizer aligning latent similarity structure with spectral views. Pretrained on approximately 80,000 hours of EEG from 29,048 subjects across 106 datasets, SPERA achieves the highest average balanced accuracy across nine downstream tasks spanning clinical, cognitive, and BCI applications. SPERA further exhibits strong parameter efficiency under linear probing and robustness across varying recording conditions, suggesting its potential as a general-purpose backbone for diverse EEG analyses.