Taming fNIRS-based BCI Input for Better Calibration and Broader Use
summary
A new multi-subject fNIRS dataset (27 recruited, 15 eligible) and open-source code accompany a 3-phase pipeline that first trains on synthetic data from other subjects, then refines on real group data, and finally fine-tunes per subject. The system classifies n-back workload from 2-second windows of oxy/deoxy signals, achieves 71% average accuracy for binary n-back versus 50% chance, and consistently outperforms baselines while using only 50% of per-user calibration data.
AI-written summary · about
chapters (12)
- 0:00Intro
- 0:24Typical BCI System
- 0:50Introduction
- 1:22ENIRS Machine and Headband
- 1:41Contributions
- 2:34Tasks
- 3:32Challenges
- 4:04Data Augmentation + Domain Adaptation
- 5:17Result 1: Better classification accuracy
- 6:09Result 2: Reducing individual calibration effort
- 6:49Conclusion
- 8:06Ongoing Effort
related talks
11mSpinning Icons: Introducing a Novel SSVEP-BCI Paradigm Based on Rotation
34mReading Thoughts with JavaScript
17mNew User Interface - Part 2
01h 02mBuilding NeuroInclusive Employee Experiences
06mmmASL: Environment-Independent ASL Gesture Recognition Using 60 GHz Millimeter-wave Signals
19mBSpeak: An Accessible Voice-based Crowdsourcing Marketplace for Low-Income Blind People
03mBiofeedback-Driven Multiplayer Games: Leveraging Social Awareness and Physiological Signals for Play
01h 26mInclusive & Ethical NLP Design Building AI for Everyone