Research map / Label-efficient and robust learning / Domain adaptation and generalization
Assessing the Impact of Population Data Domain Differences on Transfer Learning in P300-based Brain-Computer Interfaces (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Simulates P300 speller use on the bigP3BCI dataset to test whether EEGNet transfer learning from non-ALS (cross-domain) or ALS (within-domain) source data helps ALS end-users versus user-specific SWLDA and EEGNet models.
Problem
Transfer learning for BCIs is mostly studied with general-population data, but target end-users (people with ALS) differ in signal characteristics and experiment conditions; it is unclear how population data domain affects transfer learning for P300 speller use in ALS users.
Approach
Compares EEGNet models pretrained on within-domain (ALS) versus cross-domain (non-ALS) source data, with and without user-specific fine-tuning, against user-specific SWLDA, under simulated dynamic stopping with and without a language model, on a held-out ALS user.
Key results
- Generic EEGNet trained on non-ALS cross-domain source data performed poorly relative to user-specific models for most mid- to high-level ALS performers.
- EEGNet trained on within-domain ALS source data performed comparably to or better than user-specific models.
- Low-level BCI performers mostly benefited from generic EEGNet models, with the benefit depending on baseline SWLDA performance.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Pretrain EEGNet on non-ALS data, including fine-tuning on ALS or user-specific data | EEGNet for P300 speller transfer learning · ALS cohort, Study L | character accuracy | Hurt |
| Pretrain EEGNet on ALS data | EEGNet for P300 speller transfer learning · ALS cohort, Study L | character accuracy | No clear effect |
| Use cross-domain pretrained EEGNet instead of user-specific models | Mid- to high-level BCI performers · ALS cohort, Study L | character accuracy | Hurt |
| Use generic EEGNet models instead of user-specific models | Low-level BCI performers · ALS cohort, Study L | character accuracy | Helped |
Surprising findings
- Cross-domain pretraining on non-ALS data hurt most mid- to high-level performers even after fine-tuning on ALS data.
Limitations
- Small test set (single held-out ALS user, Study L) and limited statistical power.
- Only one deep learning model (EEGNet) evaluated.
- Confounds between cohorts: ALS and non-ALS studies differ in grid size and inter-stimulus interval.
Experimental setup
Datasets: bigP3BCI
Benchmarks: bigP3BCI
Metrics: character accuracy expected stopping time (stimulus presentations per character
Baselines: SWLDA, user-specific EEGNet, user-specific
Builds on: EEGNet SWLDA dynamic stopping
Closest papers by meaning
- NeuroAtlas: Benchmarking Foundation Models for Clinical EEG and Brain-Computer Interfaces (NeurIPS 2026 · similarity 0.84)
- SPINT: Spatial Permutation-Invariant Neural Transformer for Consistent Intracortical Motor Decoding (NeurIPS 2025 · similarity 0.82)
- Exploring the trade-off between deep-learning and explainable models for brain-machine interfaces (NeurIPS 2024 · similarity 0.82)
- Few-shot Algorithms for Consistent Neural Decoding (FALCON) Benchmark (NeurIPS 2024 · similarity 0.81)
- MarkovType: A Markov Decision Process Strategy for Non-Invasive Brain-Computer Interfaces Typing Systems (AAAI 2025 · similarity 0.81)
- MOTOR: A Time-to-Event Foundation Model For Structured Medical Records (ICLR 2024 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.