aiqbench analysis
This research introduces ERP-XTTN, a novel cross-attention architecture designed to make brain-computer interface classifiers both interpretable and capable of generalizing across different subjects without calibration. By routing EEG signals directly to fixed physiological prototypes, the model ensures that every classification decision is grounded in observable neurophysiological content rather than opaque attention patterns. The study demonstrates competitive performance on multiple public datasets while providing a causal, calibration-free framework for decoding diverse event-related potential components.
ERP-XTTN: Interpretable Prototype-Guided Cross-Attention for Cross-Subject ERP Classification
arXiv:2606.02939v2 Announce Type: replace Abstract: Interpretable brain-computer interface classifiers that generalize across subjects without calibration remain an open challenge. We evaluated whether prototype-based cross-attention can provide competitive, inherently interpretable ERP classification across paradigms under deployment-compatible conditions. We propose ERP-XTTN (ERP Cross-Attention), a cross-attention architecture that routes input EEG peaks to fixed difference-wave prototypes via query-key-only cross-attention with no value projection. Classification is based directly on prototype similarity and a separate measure of component amplitude, so that prototype content contributes to every decision by construction. Prototypes are derived automatically from prominent extrema in the training-fold grand-average difference wave. We evaluated across three public sources (BNCI Horizon 2020, HRI Cursor, and ERP CORE) encompassing eight ERP components (ERN, LRP, ErrP, N170, P300, N2pc, MMN, N400). Evaluations used LOSO cross-validation with causal filtering at a three-channel montage, compared against EEGNet, EEG-Deformer, EPMN, and xDAWN with Riemannian geometry. The mean performance gap between the best baseline and ERP-XTTN was 0.025 AUROC. Prototype interventions confirmed that decisions depend on prototype content rather than on the routing attention pattern alone. False positives morphologically resembled true positives more than true negatives did, so classification errors are neurophysiologically explicable. ERP-XTTN generalizes across diverse ERP morphologies under causal, calibration-free conditions, while retaining competitive performance and decisions that depend directly on physiological prototype content. Unlike post-hoc explanation methods for black-box models, the basis of each decision is directly observable in the trained model. To our knowledge, this is the first epoch-level LOSO benchmark on ERP CORE.
Abstract:Interpretable brain-computer interface classifiers that generalize across subjects without calibration remain an open challenge. We evaluated whether prototype-based cross-attention can provide competitive, inherently interpretable ERP classification across paradigms under deployment-compatible conditions. We propose ERP-XTTN (ERP Cross-Attention), a cross-attention architecture that routes input EEG peaks to fixed difference-wave prototypes via query-key-only cross-attention with no value projection. Classification is based directly on prototype similarity and a separate measure of component amplitude, so that prototype content contributes to every decision by construction. Prototypes are derived automatically from prominent extrema in the training-fold grand-average difference wave. We evaluated across three public sources (BNCI Horizon 2020, HRI Cursor, and ERP CORE) encompassing eight ERP components (ERN, LRP, ErrP, N170, P300, N2pc, MMN, N400). Evaluations used LOSO cross-validation with causal filtering at a three-channel montage, compared against EEGNet, EEG-Deformer, EPMN, and xDAWN with Riemannian geometry. The mean performance gap between the best baseline and ERP-XTTN was 0.025 AUROC. Prototype interventions confirmed that decisions depend on prototype content rather than on the routing attention pattern alone. False positives morphologically resembled true positives more than true negatives did, so classification errors are neurophysiologically explicable. ERP-XTTN generalizes across diverse ERP morphologies under causal, calibration-free conditions, while retaining competitive performance and decisions that depend directly on physiological prototype content. Unlike post-hoc explanation methods for black-box models, the basis of each decision is directly observable in the trained model. To our knowledge, this is the first epoch-level LOSO benchmark on ERP CORE.
Accepted for publication in Journal of Neural Engineering. 39 pages including supplementary material, 7 figures, 24 tables. v2: accepted manuscript replacing the preprint
Machine Learning (cs.LG); Signal Processing (eess.SP)
arXiv:2606.02939 [cs.LG]
(or arXiv:2606.02939v2 [cs.LG] for this version)
https://doi.org/10.1088/1741-2552/ae9f95 DOI(s) linking to related resources
Full text under the publisher's open licence · 284 words
Related coverage
ERP-XTTN: Interpretable Prototype-Guided Cross-Attention for Cross-Subject ERP Classification
arXiv · Jun 3, 2026
The J-Space Debate, Agent Swarms, and Pacing Frontier AI - Digital Minds Newsletter #4
LessWrong · Sep 18, 2026
Neural Visual Decoding via Cognitive guided Adaptive Blurring and Information Constrained Alignment
arXiv · Oct 17, 2026
The Continual Constitutional Convention: How language models revise the documents that govern them
AI Alignment Forum · Oct 13, 2026