ARTFEED — Contemporary Art Intelligence

Position-Adaptive Time Scheduling for EEG Generation

ai-technology · 2026-08-04

A new framework for EEG generation, built on conditional flow matching, introduces Position-Adaptive Time Scheduling to address the heterogeneity of EEG moments. The method tracks per-position reconstruction error to modulate time progression, incorporates Factorized Spatio-Temporal Attention, and uses a frequency-aligned multi-resolution spectral consistency loss. This approach aims to improve EEG generation for brain-computer interface applications by better modeling inter-channel dependencies and compensating for spectral bias.

Key facts

  • EEG generation is essential for alleviating data scarcity in brain-computer interface applications.
  • Existing flow-based approaches assume a single global time progression for all channels and time segments.
  • The proposed framework uses conditional flow matching.
  • Position-Adaptive Time Scheduling modulates position-specific time progress based on reconstruction error.
  • Factorized Spatio-Temporal Attention models inter-channel dependencies induced by volume conduction.
  • A frequency-aligned multi-resolution spectral consistency loss compensates for the power law spectral bias of EEG.
  • The framework is introduced in arXiv paper 2608.00048.
  • The paper was announced as a cross-type on arXiv.

Entities

Institutions

  • arXiv

Sources