The Identity Trap in EEG Foundation Models: A Diagnostic Audit

2026-06-04

AI Summary

This study addresses a critical yet previously unrecognized issue in EEG foundation models: their high cross-subject performance may stem not from genuine clinical biomarkers but from reliance on subject identity cues—a phenomenon the authors term the “identity trap.” To systematically diagnose this, the work introduces FMScope, a novel evaluation protocol comprising five techniques: variance decomposition under frozen representations, subject-axis erasure, aperiodic 1/f component ablation, hierarchical label probing, and within-subject directional consistency analysis. Findings reveal that identity information dominates all tested models, with effect sizes 13–89 times stronger than random baselines. Erasing identity cues boosts task decoding accuracy by up to 27 percentage points. The aperiodic 1/f signal serves as a key carrier of identity in some models, and fine-tuning enhances task-relevant signals only when consensus biomarkers are present.

Abstract

Objective. EEG foundation models (FMs) report strong accuracy on clinical resting-state EEG. However, high accuracy under subject-disjoint cross-validation remains ambiguous: it can reflect a genuine clinical biomarker, or subject-identity features that correlate with the label. We name this the Identity Trap and ask whether it can be diagnosed at the representation level before fine-tuning. Approach. We propose FMScope, a frozen-representation protocol packaging five diagnostics: variance decomposition, subject-axis erasure, aperiodic 1/f ablation, layer-wise label probing, and within-subject direction consistency. We apply it to three pretrained FMs (LaBraM, CBraMod, REVE) across four datasets in a 2x2 layout: subject relation of label x presence of a consensus cross-subject EEG marker. Main results. (i) The Identity Trap is universal: frozen subject-variance is 13-89x a random null in 12/12 pairs, rising in all 12 under fine-tuning (+10 to +63 pp). This dominance is a removable linear axis: erasing it improves label decoding where the label varies within subject (+6 to +12 pp in primary cells; +4 to +27 pp across external cohorts). (ii) Aperiodic 1/f is one subject carrier: removing it drops the subject probe by 9-19 pp on LaBraM and CBraMod. REVE saturates subject identity without measurable aperiodic dependence. (iii) Fine-tuning amplifies label-variance only in cells with a literature-established cross-subject marker. Significance. The Identity Trap is a physically-grounded instance of shortcut learning: the preferred cue has a measurable physiological component, and subject-disjoint splitting alone cannot rule it out. FMScope separates gains reflecting a biological marker from those reflecting subject identity.

Problem

• Identity Trap
• EEG foundation models
• subject identity
• shortcut learning
• clinical biomarker

Innovation

• Identity Trap
• EEG foundation models
• shortcut learning
• FMScope
• subject identity