Transfer of hierarchical EEG alignment to MEG

Determine whether the hierarchical alignment framework for EEG-based visual retrieval and reconstruction transfers to magnetoencephalography (MEG), including whether the neural-visible optimal layer changes because of differences in source sensitivity and spatial resolution.

Background

The paper proposes a hierarchical alignment framework that maps EEG signals to intermediate CLIP representations selected for their neural visibility, and uses these representations for visual retrieval and image reconstruction. The authors note that the framework was evaluated only with EEG recordings from the THINGS-EEG dataset.

MEG differs from EEG in source sensitivity and spatial resolution. Consequently, the CLIP layer that is most reliably aligned with neural signals may not be the same across the two modalities. Establishing whether the framework generalizes to MEG, and whether its neural-visible optimal layer shifts, requires cross-modal and cross-dataset validation.

References

Third, it is unclear whether the hierarchical alignment framework transfers to other modalities such as MEG. Because MEG differs from EEG in source sensitivity and spatial resolution, the neural-visible optimal layer may shift across modalities; cross-dataset and cross-modal validation will be needed.

— EEG-based Visual Retrieval and Reconstruction: From Neurally Visible Optimal Layer to Hierarchical Diffusion Generation  (2609.02582 - Wang et al., 2 Sep 2026) in Section Discussion, paragraph beginning “This study has several limitations.”