Cross-Subject Statistical Shift Estimation for Generalized Electroencephalography-based Mental Workload Assessment
archive: archived pipeline: cataloged
Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)
Summary
**Research Question and Motivation** This paper addresses the challenge of cross-subject variability in EEG-based mental workload assessment, which hinders the generalization of passive brain-computer interface (BCI) models to new users. While domain adaptation (DA) strategies are often employed to mitigate this issue, they typically rely on assumptions—such as covariate shift—that may not hold for EEG data, where conditional label distributions $p(y|x)$ often shift between subjects. The authors propose a method to explicitly estimate both marginal and conditional statistical shifts between subjects to validate these assumptions and guide the selection of appropriate DA strategies. **Methods and Experimental Design** The study utilizes the WAUC dataset, comprising EEG recordings from 18 subjects performing mental tasks while engaging in physical activity (either running on a treadmill or pedaling a stationary bike). The authors develop two estimation strategies: one for conditional shift, using a k-nearest neighbor labeling function to approximate the mismatch between subject-specific labeling rules, and another for marginal shift, using a Random Forest classifier to estimate the H-divergence between subject feature distributions. These estimates are aggregated into Hermitian disparity matrices. The experiments compare four feature normalization strategies: no normalization, per-subject whitening (z-score), and normalization relative to two different baseline periods (one with no physical activity, one with physical activity only). Cross-subject performance is evaluated using a leave-one-subject-out (LOSO) cross-validation scheme with Random Forest classifiers. **Findings** The results demonstrate that feature normalization significantly impacts estimated statistical shifts. Per-subject whitening consistently reduced the aggregate conditional shift for both treadmill and bike conditions, aligning with its known benefits for classification performance. However, baseline-dependent normalization showed divergent effects: for the bike condition, it yielded only slight improvements in conditional shift, whereas for the treadmill condition, normalizing against the first baseline (no physical activity) actually increased the estimated conditional shift, potentially degrading model generalization. Conversely, normalizing against the second baseline (physical activity only) reduced the shift to levels comparable to whitening. The analysis revealed that the bike condition exhibited higher cross-subject conditional shift than the treadmill condition, indicating that EEG responses during cycling are more subject-specific. Marginal shift estimates, derived from pairwise subject classification accuracy, were also affected by normalization, with whitening generally reducing the divergence between subjects. **Significance** This work provides a quantitative framework for diagnosing distributional shifts in EEG data, moving beyond heuristic normalization practices. By linking specific normalization strategies to measurable changes in conditional and marginal shifts, the study offers actionable insights for designing robust, cross-subject BCI systems. It highlights that the effectiveness of standard normalization techniques is context-dependent, varying significantly with the type of physical activity, and underscores the necessity of verifying DA assumptions before applying them to real-world mental workload monitoring applications.
Provenance
The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | Crossref | — | — | 1 | 2026-08-09 |
| archive | success | unpaywall | — | — | 2 | 2026-08-09 |
| extract | success | cached | — | — | 4 | 2026-08-23 |
| clean | success | clean | — | — | 1 | 2026-08-09 |
| chunk | success | chunk | — | — | 1 | 2026-08-09 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-08-09 |
| promote | success | — | — | — | 1 | 2026-08-09 |
| summarize | success | llm | qwen3.8-27b-gittensor | summ-v5 | 3 | 2026-08-23 |
| tag | success | vector_similarity | — | — | 11 | 2026-08-11 |
| verify | success | — | — | — | 2 | 2026-08-09 |
Summary generated by qwen3.8-27b-gittensor on 2026-08-23; verification: pending re-verification.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.
Information type
What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).
- Empirical Findings: physiological data