Research map / Label-efficient and robust learning / Recognition and long-tail
L-FAME: Longitudinal Focused Attention Meditation EEG Dataset and Benchmark
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.
Introduces L-FAME, a 6-week longitudinal 64-channel EEG dataset from 74 college students practicing three focused attention meditation techniques (Breath Focus, Hare Krishna, SA-TA-NA-MA), with a three-task ML/DL benchmark covering state decoding, technique classification, and cross-session adaptation.
Problem
Meditation EEG decoding research lacks longitudinal, multi-technique datasets and standardized benchmarks, and existing models fail to generalize across subjects and over time.
Approach
A longitudinal EEG dataset with pre/post six-week training across three focused attention meditation paradigms, paired with a benchmark of intra-subject, inter-subject, LOSO, technique classification, and zero/few-shot cross-session adaptation tasks.
Key results
- Task 1 intra-subject AUC reaches 99.2% (EEGNet) but inter-subject and LOSO AUCs fall to roughly 66-70%.
- Task 2 cross-subject technique classification PR-AUC stays low, e.g., EEGNet 35.0% pre and 48.8% post vs 33.3% chance.
- Task 3 zero-shot AUC drops to 63-67%, and full fine-tuning with 30 post-intervention samples recovers EEGNet to 79.3%.
Surprising findings
- Traditional PSD and FBCSP with SVM outperform deep models in intra-subject evaluation but underperform them in cross-subject and LOSO settings.
- Cross-subject technique classification remains near chance even after six weeks of training.
Limitations
- Attrition reduced the longitudinal cohort from 74 to 44 participants.
- Lack of a control group, reliance on self-reported adherence, and two sparse time points.
- Restricted college-aged demographic limits clinical generalizability.
Experimental setup
Datasets: L-FAME
Benchmarks: L-FAME
Metrics: AUC PR-AUC
Baselines: PSD + SVM FBCSP + SVM ShallowConvNet DeepConvNet EEGNet EEG-Conformer
Builds on: EEGNet EEG-Conformer BIDS-EEG
Closest papers by meaning
- NeuroAtlas: Benchmarking Foundation Models for Clinical EEG and Brain-Computer Interfaces (NeurIPS 2026 · similarity 0.85)
- What do EEG Foundation Models Capture from Human Brain Signals? (NeurIPS 2026 · similarity 0.83)
- EEG Benchmarking Needs a Task Specification Layer: NeuroDoc for Rulebook-Guided, Executable Benchmark Construction (NeurIPS 2026 · similarity 0.82)
- FEEL: Quantifying Heterogeneity in Physiological Signals for Generalizable Emotion Recognition (NeurIPS 2025 · similarity 0.82)
- BrainLM: A foundation model for brain activity recordings (ICLR 2024 · similarity 0.82)
- Is Limited Participant Diversity Impeding EEG-based Machine Learning? (NeurIPS 2025 · similarity 0.82)
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.