Standardized benchmark pipeline for reproducible cross-subject and within-subject classification
The technique defines an extensible benchmark for fNIRS classification that includes standardized benchmark tasks (cross-subject and within-subject), and a fully automated pipeline from raw data to evaluation to enable reproducible comparison across methods and datasets.
Discovered 2026-11-10 · Sources checked 2026-11-10
Why it matters
It addresses limited, inconsistent datasets and evaluation protocols by providing uniform task definitions and an automated evaluation pipeline, which helps reveal when methods (including those validated on limited data) fail to generalize to new evaluation settings.
What changed
Instead of relying on small, isolated evaluations, the approach aggregates many public fNIRS data collections into a single benchmark with consistent preprocessing-to-evaluation automation and standardized task protocols.
Where it helps
- Comparing ML classification methods on fNIRS under consistent cross-subject protocols
- Evaluating within-subject fNIRS classification with standardized tasks
- Running reproducible model evaluations across multiple fNIRS experimental paradigms
- Studying generalization of time-series or neuroimaging classifiers on diverse task settings
- Building extensible benchmarks for other biosignal modalities with similar evaluation needs
Evidence & reproduction
Reproduction not evaluated. A linked paper or repository does not establish independent reproduction.
Limitations & open questions
The provided text does not specify model architectures, training details, or quantitative performance characteristics; it only reports that performance can vary across tasks and that some methods may not generalize.
Related World Models papers
Sources
- fNIRSAtlas: A Large-Scale Benchmark for Functional Near-Infrared Spectroscopy Classification ↗
arXiv · primary · Checked 2026-11-10
