#25 · Systems biology

Distributional and cross-validated distances for comparing neural representations

The dominant thread is how neural representations are compared, specifically the choice of distance, comparison metric and inferential procedure. A Wasserstein-distance framework applied to EEG quantifies representational stability and dimensionality across aging, mild cognitive impairment and Alzheimer's disease. Separately, generalized cross-validated Euclidean and correlation distances are proposed for multivariate pattern analysis to improve reliability, and null-distribution randomization schemes for neural tracking correlations are shown to encode different hypotheses, with a Fisher-transform approach proposed. The published EEG item compares canonical neural population models and uses grammar-based generation to fit candidate architectures to EEG spectra, addressing model structure rather than distance directly. What is new is the shift from single summary values toward distributional measures and explicit hypothesis specification. Evidence for the distance and inference work remains preprint-based.