Conference on Cognitive Computational Neuroscience

The representational geometry of a brain region can be characterized by the distances among neural activity patterns for a set of experimental conditions. Researchers routinely estimate representational distances from brain-activity measurements that either sparsely sample the underlying neural population (e.g. neural recordings) or pool across the activity of many neurons (e.g. fMRI voxels). Here we use theory and simulations to clarify under what circumstances representational distances estimated from brain-activity measurements reflect the representational geometry of the underlying neural population, and what distortions must be expected under other circumstances. We demonstrate that if single neurons are sampled at random, the estimated representational distances are undistorted. For voxels that take nonnegatively weighted averages, the resulting geometry overemphasizes the population-mean dimension, while accurately capturing orthogonal dimensions. Removing the mean from voxel patterns recovers the underlying representational geometry exactly under idealized conditions. This explains why the correlation distance, the most popular measure of representational dissimilarity, “works” so well, yielding geometries that can appear similar between fMRI and neural recordings.