A new research paper published on arXiv explores the challenges of using functional near-infrared spectroscopy (fNIRS) for autism spectrum disorder (ASD) classification. The study introduces a cross-time-window transfer benchmark to address performance degradation caused by variations in optimal observation windows across subjects. Findings indicate that while zero-shot classification accuracy is low, minimal subject-specific fine-tuning significantly improves results, highlighting inter-subject variability as a key barrier. The research also demonstrates the effectiveness of domain-adversarial and self-supervised strategies in achieving robust classification without target-subject data, suggesting that relevant information can be extracted from short fNIRS windows. AI
RANK_REASON Academic paper published on arXiv detailing a new benchmark for fNIRS-based autism classification. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
