MontageFlow: Physically Valid High Fidelity EEG Generation
Generative modeling offers a promising solution to the scarcity of labeled clinical electroencephalography (EEG) data under strict privacy constraints. Although recent generative models report substantial improvements in generation fidelity, we show that these gains conceal a categorical failure.
arXiv - cs.AI2 min read