# Matched-Input Estimates Differ in Sign Across Architectures: Auditing EEG Foundation Models on Motor Imagery

arXiv: https://arxiv.org/abs/2609.23924 · HTML: https://mlsanity.com/paper/2609.23924/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 13 of its 28 references that are in the map.
- First posted on arXiv 20 Sep 2026.

## Builds on (most cited first, 13 of 13)

- [On Calibration of Modern Neural Networks](https://mlsanity.com/paper/1706.04599/) (2017)
- [EEGNet: A Compact Convolutional Network for EEG-based Brain-Computer Interfaces](https://mlsanity.com/paper/1611.08024/) (2016)
- [Large Brain Model for Learning Generic Representations with Tremendous EEG Data in BCI](https://mlsanity.com/paper/2405.18765/) (2024)
- [CBraMod: A Criss-Cross Brain Foundation Model for EEG Decoding](https://mlsanity.com/paper/2412.07236/) (2024)
- [MOABB: Trustworthy algorithm benchmarking for BCIs](https://mlsanity.com/paper/1805.06427/) (2018)
- [The largest EEG-based BCI reproducibility study for open science: the MOABB benchmark](https://mlsanity.com/paper/2404.15319/) (2024)
- [EEG-FM-Bench: A Comprehensive Benchmark for the Systematic Evaluation and Diagnostic Analyses of EEG Foundation Models](https://mlsanity.com/paper/2508.17742/) (2025)
- [AdaBrain-Bench: Benchmarking Brain Foundation Models for Brain-Computer Interface Applications](https://mlsanity.com/paper/2507.09882/) (2025)
- [NeuralBench: A Unifying Framework to Benchmark NeuroAI Models](https://mlsanity.com/paper/2605.08495/) (2026)
- [OmniEEG-Bench: A Standardized Evaluation Benchmark for EEG Foundation Models](https://mlsanity.com/paper/2606.00815/) (2026)
- [Brain4FMs: A Benchmark of Foundation Models for Electrical Brain Signal](https://mlsanity.com/paper/2602.11558/) (2026)
- [NeuroAtlas: Benchmarking Foundation Models for Clinical EEG and Brain-Computer Interfaces](https://mlsanity.com/paper/2605.14698/) (2026)
- [A Negative-Control Protocol for Clinical EEG Foundation-Model Benchmarks: Dataset Identity and External-Cohort Stress Testing](https://mlsanity.com/paper/2607.24519/) (2026)

## Abstract (arXiv)

Pretrained EEG foundation models are increasingly proposed as general-purpose encoders for brain-computer interfaces, yet recent benchmarks disagree about when their representations transfer to downstream tasks. We audit LaBraM and CBraMod on motor imagery under a validation-locked protocol in which preprocessing, architecture, optimization, freeze depth, checkpoint, temperature, and method selection are determined using training-session data only. On four-class BCI Competition IV-2a, every supervised comparator evaluated here outperforms every foundation-model configuration, including validation-selected fine-tuning. We then examine a key confound: foundation models and task-specific decoders are normally evaluated using different input pipelines. Retraining three supervised architectures on the broadband arrays consumed by the foundation models produces matched-input accuracy differences of opposite sign across architectures: broadband input improves ATCNet by 0.078 accuracy while reducing EEG Conformer accuracy by 0.088. None of the three individual matched-input terms is significant after multiple-comparison correction at n = 9, so we treat the sign variation descriptively rather than as a formal architecture-by-pipeline interaction. These observed sign differences suggest that a single comparator may not provide an architecture-invariant decomposition of a pretrained-versus-supervised performance gap. The four-class deficit also does not reproduce uniformly across motor-imagery datasets: on two-class BNCI2014-004 we cannot detect the same separation between fine-tuned CBraMod and the supervised comparators. Finally, validation-fitted temperature scaling returns foundation-model calibration error to the supervised range despite substantially lower four-class accuracy.