NeuroShield: A Device-Agnostic Foundation Model for EEG Authentication

NeuroShield is a foundation model that enables EEG-based biometric authentication across different hardware devices and recording configurations. The model was pretrained on over 15,000 subjects and demonstrates significant accuracy improvements while generalizing to unseen equipment and data formats.

NeuroShield addresses a fundamental fragmentation problem in EEG biometric research where authentication models remain locked to specific hardware configurations and data collection protocols. Traditional approaches treat each new headset or dataset as an independent problem, preventing knowledge transfer and forcing repeated model development cycles. This paper presents a transformer-based architecture that ingests variable-channel and variable-length EEG recordings, learning identity-discriminative embeddings that transcend hardware constraints. The dual-stage transformer design enables the model to handle heterogeneous input formats while maintaining authentication performance. Pretrained on three public datasets encompassing 15,762 subjects and 28,116 sessions, NeuroShield demonstrates transfer learning capabilities on downstream tasks with error rate reductions of 0.44-8.06 percentage points compared to existing approaches. The model generalizes beyond its training distribution, functioning on longer segments and unfamiliar channel layouts, establishing practical reusability across real-world deployment scenarios. For the biometric authentication industry, NeuroShield accelerates adoption of EEG-based identity verification by reducing implementation friction. Developers no longer require custom models per hardware variant, lowering deployment costs and enabling faster product iteration. The open-source release multiplies adoption velocity through community contributions and integration into authentication systems. However, EEG authentication remains nascent for commercial scaling; widespread adoption requires solving practical challenges like user comfort, real-time processing requirements, and competitive advantages over existing biometric modalities. The foundation model approach demonstrates the viability of transfer learning in biometric domains, potentially influencing how multimodal authentication systems are developed.

• →NeuroShield eliminates hardware-specific model constraints through a device-agnostic transformer architecture handling variable EEG inputs.
• →Pretraining on 15,762 subjects enables superior transfer learning with measurable error reductions on unseen datasets.
• →The model generalizes to longer segments and unfamiliar channel configurations beyond training distribution specifications.
• →Open-source release accelerates commercial EEG authentication adoption by reducing per-device development overhead.
• →Foundation model approach validates transfer learning applicability in biometric identity verification across heterogeneous hardware.