Figure 1.

Overview of the different multivariate analyses. (A) Neural representations can be understood as points in an n-dimensional space, where n corresponds to the number of spatial units considered. A two-dimensional (i.e., two voxels) space is shown here for visualization purposes. Each condition (in this example, each stimulus) is defined by its associated activation across the spatial units. (B) Decoding analysis seeks to assess how differentiable are the conditions' representations in this space. To do so, a boundary (shown in red) that separates two (or more) conditions is first estimated and later tested with independent, unlabeled activity patterns. In this example, the boundary classifies between two category conditions: animate and inanimate stimuli. (C) Representational Similarity Analysis (RSA) describes the geometry of the space computing the similarity structure across conditions. In the example, the estimated similarity matrix shows that the representations are organized not only by the stimulus category (animate and inanimate stimuli, lower and upper quadrants shown in green) but also by the animate stimulus subcategory (mammals and non-mammal animals, shown in purple). RSA also enables characterizing the estimated encoding space using theoretical models. In this case, the third model (which predicts a broader animate/inanimate distinction together with further animate subcategory differentiation) will better predict brain data. (D) Canonical Template Tracking (CTT) estimates specific canonical neural representations using localizer tasks (lower panels) and then assesses their activation level during the main paradigm (upper panel). In the example, the paradigm is a memory retrieval task in which the participants retrieve the images that were associated with each word in a previous learning phase. The stimuli sensory templates (Localizer 1) are estimated with a task where upside-down images need to be detected, while semantic templates (Localizer 2) are estimated with a task in which the stimuli are categorized as mammals or non-mammals. Those templates are then compared against the activity patterns associated with the items retrieved during the main paradigm. Critically, the activation strength from the retrieved image template (in the example, the cat, shown in red) can be compared against the non-retrieved ones (e.g., the dog templates), which can act as baseline condition. Cat., Category; Subcat., Subcategory. Stimulus images were retrieved from the stimulus database employed in González-García et al. (2021), CCBY 4.0.