NeuroAdapt-Bench: TTA methods fail on EEG foundation models — optimization-free methods more stable

NeuroAdapt-Bench, a benchmark for test-time adaptation (TTA) on EEG foundation models, reveals that standard TTA methods yield inconsistent gains and often degrade performance under distribution shifts across clinical settings, devices, and populations. Gradient-based approaches suffer heavy degradation, while optimization-free methods show greater stability and reliable improvements. The findings highlight the need for domain-specific adaptation strategies in EEG.

• #eeg
• #foundation-model
• #test-time-adaptation
• #benchmark
• #distribution-shift