2.1.3. Procedure

Before coming to the scanning center, the subjects were pre-screened over the telephone to ensure eligibility. Upon arrival at the center, a structural T1 scan was acquired, followed by an exposure block with simultaneous scanning. During this block, the subjects were asked to passively look at the screen and refrain from pressing any buttons. No other context or instructions were given. Each of the nine training formations were shown one at a time for 2 s each before a fixed 10 s ITI. Each formation was shown four times for 36 total trials. Following the exposure block, subjects were read instructions for the experiment. They were told how many formations there would be in the experiment, along with the set number of categories the formations belonged to. Their job was to figure out which formations belong to each category by pressing the buttons on their response pad and utilizing the corrective feedback. A brief practice test (un-scanned) was given, where they learned to categorize unrelated formations. After practice, six training blocks were run with brief breaks in-between and, after training, the subjects sat through another exposure block where they passively viewed each stimulus. To end the experiment, the subjects went through the generalization block, given only the instructions that they were going to go through a final block with no feedback. They were not told whether there would be novel formations in this block. The subjects were asked to write-down their strategies in a debrief questionnaire for categorizing the formations before receiving compensation and leaving the center.

#### 2.1.4. fMRI Acquisition and Pre-Processing

MRI data were acquired with a 3T Siemens Skyra. A high-resolution T1-weighted MPRAGE was acquired for co-registration and normalization before the task was administered (TR = 2.5 s, TE = 3.41 ms, flip angle = 7◦, matrix size = 256 × 256, FOV = 256 mm, 1 mm isotropic). Whole-brain fMRI was acquired using a gradient-echo EPI pulse sequence: TR = 2 s, TE = 26 ms, 100 × 100 matrix, FOV = 200 mm, 72 oblique axial slices, no skip, 2 mm isotropic voxels, GRAPA factor 2, multiband factor 3. Preprocessing was carried out in FSL version 5.0.9 (www.fmrib.ox.ac.uk/fsl). The functional images were skull stripped using BET (brain extraction tool), motion corrected, co-registered to the T1 anatomical image, and smoothed with a 4mm FWHM kernel. Two sets of analyses were performed: traditional whole-brain univariate analyses and trial-by-trial multivoxel pattern classification analyses within a set of a priori ROIs. For univariate whole-brain group analyses, functional data were registered to standard MNI space. For multivoxel pattern classification, the data were kept in native space of each participant and trial-specific activation patterns (betaseries) for classification analysis within each participant were extracted using a general linear model with a separate regressor for each trial [25]. A detailed description of each of these analysis approached is provided together with results for better readability.
