Caplette & Lippé's arXiv paper boosts neural distance metrics for MEG pattern analysis

Multivariate pattern analysis (MVPA) is a cornerstone of cognitive neuroscience, allowing researchers to measure how neural representations differ between experimental conditions. Prior work by Guggenmos et al. (2018) recommended the cross-validated Euclidean distance or the within-class-corrected Pearson distance for MEG data. In this new arXiv preprint (2608.10394), Laurent Caplette and Sarah Lippé show these distances can be improved. They first demonstrate that the cross-validated Euclidean distance is mathematically equivalent to a sum of between-partition distances, a relationship that yields a generalized variant with higher reliability and accuracy.

Second, the authors leverage the formal link between Euclidean distance and Pearson correlation to define a cross-validated correlation distance. This new measure is both more accurate and more interpretable than the one proposed by Guggenmos and colleagues. The paper also compares generalized cross-validation against within-class correction, showing that the generalized approach achieves higher accuracy for correlation distances. These refinements give MEG researchers more robust tools for tracking the temporal dynamics of neural representations, which is critical for understanding perception, memory, and decision-making.