This page presents the raw data described in this paper:

• DA Kahn*, MG Mattar*, GK Aguirre (Submitted) “A single mechanism of temporal integration unites
neural adaptation and norm-based coding”

Please refer any questions to aguirreg@mail.med.upenn.edu.

Neural responses are modulated both by the immediate effect of stimulus similarity (neural adaptation) and the central tendency of long-term sensory experience (norm-based coding). We tested the hypothesis that both are manifestations of a single mechanism that integrates over intermediate timescales. We demonstrate a gradient of stimulus temporal integration across the human visual cortex that unifies neural and behavioral studies of prototype and similarity.

Stimuli for the experiment were generated using the GenHead software. The set of stimuli was defined by principally by 3 axes, with 3 points along each axis, resulting in 27 faces. There was a fourth axis for the stimulus set, rendering the faces older and younger appearing.

Two sets of stimuli were created and used in two different experiments, the “MDS” faces and the “RGI” faces.

Following scanning, subjects provided explicit measures of perceptual similarity for the stimuli. These measures were entered into a multidimensional scaling analysis, yielding coordinates of the perceptual stimulus spaces used as the basis for modeling in the fMRI analyses. The resulting coordinates are below:

Dataset #1 (“RGI”)

Dataset #2 (“MDS”)

20 subjects were studied in Dataset #1 (“RGI”), five of whom were discarded for head motion, missing data, or poor performance on the scanner cover task. 21 subjects were studied in Dataset #2 (“MDS”), two of whom were discarded for head motion.

During scanning, subjects performed an attention cover-task. On each trial the older or younger version of each face was randomly presented. Subjects were asked to make a bilateral button press (inner two buttons on a linear button response pad) with their thumbs in response to older faces, and a different bilateral button press (outer) for younger faces. Performance data for this cover task were recorded. The primary measure of interest was the percentage of trials for which the subject failed to make a response. This score was taken as an index of inattention, and subjects who missed more than 15% of the trials were discarded. Accuracy in performance of this demanding discrimination was also calculated and is reported in the table below.

Performance and demographic information for the different subjects are:

Dataset #1 - “RGI”

Dataset #2 - “MDS”

The stimuli were presented in an order defined by a Type 1, Index 1, k=28, first-order counterbalanced sequence. The resulting cyclical sequence is 784 elements long. The “zero label” trials were doubled at each occurrence, yielding a sequence with 812 elements. Two complete sequences were presented to each subject, for a total of 1624 “trials”. Optimal sequences were selected based upon an assumption of linear adaptation by distance in the face space. Each subject was shown two different sequences which were concatenated. The two sequences were selected to have the same starting label to allow this concatenation.

Each subject in the Dataset #1 (“RGI”) was shown the stimuli using the same sequence. For Dataset #2 (“MDS”), each subject was shown the stimuli guided by a different sequence.

Each trial was 1.5 seconds long, and the TR of image acquisition was 3 seconds, resulting in 812 imaging data-points per subject. The imaging data were collected as six separate scans. This required breaking the sequence into pieces. The stimuli presented at the start of each scan would therefore lack the “context” that produces the carry-over effect, as they would not have been immediately proceed by other stimuli. To account for this, at the start of each scan the last ten stimuli from the prior scan were first presented (i.e., 5 TRs worth of data). For the first scan, the last 10 stimuli from the entire sequence were presented, completing the cycle.

Consequently, each scan collected 141 TRs. The first 5 TRs of each scan were discarded in pre-processing. Additionally, the final 4 TRs of the last scan were discarded, as there were no more stimuli left to present for those final image time-points. This yielded 136 images from each of 5 scans, and 132 images from the final scan, giving a total of 812 images in the final data.

We have made available for download a .zip file with all MATLAB scripts used in the analysis. Alternatively, each individual script can be read directly on the browser through the following links.

The following MATLAB scripts were used to generate the drift covariates, create the ROI, and run GLMs on the surface space.

• 1_prepData_create_all_driftCovs.m
• 2a_prepData_run_Drift_glms.m
• 2b_prepData_run_MainEffect_glm.m
• 2c_prepData_adaptPlusID_outputAdapt_glm.m
• 2d_prepData_adaptPlusID_outputID_glm.m
• 3_prepData_create_mask.m

The following scripts correspond to the main analyses of the paper.

Each of script below corresponds to one of the figures from the paper. In general, they build the figures directly from the raw surface data, although some of the functions from the libraries below may be necessary.

• Figure 2C - Traditional Discrete Models - ROI analysis
• Figure 2D - Drifting Norm Model, ROI analysis
• Figure 3A/C - Cortical Surface Maps
• Figure 3B/D - Measured EC along A-P axis

Some of the additional analyses presented in the paper were conducted with the following scripts.

• Model Comparison
• Equivalence to exponential models
• Hemisphere comparison

• driftCov.m
• auxFunctions

The plots below show rotational head movement during the concatenated fMRI scans of the experiment. These subjects were discarded because of continuous pitch (yes-yes) movements of the head during scanning. This motion was the result of padding on the sides of the head preventing yaw and roll movements, but no padding to the forehead to prevent pitch rotations. The X-axis is in units of TRs, and the Y-axis is rotation in degrees. The color code used is:

Back to Subjects Overview

Back to Subjects Overview

Back to Subjects Overview

Back to Subjects Overview

Back to Subjects Overview