Schedule of Events | Symposia

Graduate Student Award Winner

Trial-level representational similarity analysis

Poster Session F - Tuesday, April 1, 2025, 8:00 – 10:00 am EDT, Back Bay Ballroom/Republic Ballroom

Shenyang Huang¹ (shenyang.huang@duke.edu), Cortney M. Howard¹, Ricardo Morales-Torres¹, Matthew Slayton¹, Paul C. Bogdan¹, Roberto Cabeza¹, Simon W. Davis¹; ¹Duke University

A neural representation refers to the brain activity that stands in for one’s cognitive experience, such as seeing an image of a cardinal. In cognitive neuroscience, one major approach to studying neural representation is representational similarity analysis (RSA). The classic RSA (cRSA) method examines the overall quality of representations across a number of instances (e.g., cardinal, broccoli, hammer) by assessing the correspondence between two representational dissimilarity matrices (RDMs): one based on some theoretical model of stimulus similarity and the other based on neural activity data. However, cRSA fails to appropriately account for three levels of statistical variability: participant, stimulus, and trial. Here we formally introduce trial-level RSA (tRSA), an innovative analytical framework that aims to measure the strength of neural representation for a single instance. First, we demonstrated how tRSA works by comparing RDMs at the level of trials, and we validated its numerical correspondence to cRSA in quantifying overall representation strength. Second, we compared the statistical inferences drawn from both frameworks using simulated data that reflected a wide range of scenarios. Modeling tRSA in a multi-level framework was more theoretically appropriate and demonstrated significant enhancements in Type I and Type II error rates compared to cRSA. Third, using a real fMRI dataset, we further demonstrated the issues with cRSA, validated the robustness of tRSA, and presented some novel analyses of neural representation that are not answerable with cRSA but feasible with tRSA. In summary, tRSA proves to be a robust and versatile analytical approach for cognitive neuroscience at large.

Topic Area: METHODS: Other