Schedule of Events | Search Abstracts | Invited Symposia | Symposia | Poster Sessions | Data Blitz

Classifying stimulus-level memory outcome using temporal versus spectrographic features

Poster Session E - Monday, March 9, 2026, 2:30 – 4:30 pm PDT, Fairview/Kitsilano Ballroom

Kieran Pawluk¹ (), Angad Chahil¹, Faisal Anqouor¹, Jeremy Caplan¹; ¹University of Alberta

An exciting recent trend in memory research is to use machine-learning-style classifiers to identify brain activity that is in some sense predictive of memory outcome at the stimulus level, rather than just descriptive. For example, brain activity during processing of a recognition probe item can be used to classify whether a target-probe trial is about to be a hit or a miss. This approach has been taken with fMRI and spectrographic (power as a function of time, frequency and electrode) features but rarely (Chakravarty et al., 2020), with the time-domain signal (voltage as a function of time and electrode). Danyluik et al. (in press), in a feedback-driven learning task, found spectrographic features far outperformed time-domain features, despite the fact that spectrographic features should have access to the same underlying information present in the time domain. Asking if the spectrographic advantage is always the case, we conducted the same kind of comparison in the test phase of an old/new verbal item-recognition dataset. Spectrographic features nominally outperformed time-domain features when classifying hits versus misses, but the advantage was not significant, despite a large sample (N=63). Hypotheses about the cause of the superiority like nonlinearity and temporal smoothing were not supported. In sum, memory classification should continue to use spectrographic features, but time-domain features hold comparable promise and have some unique advantages such as greater temporal specificity.

Topic Area: METHODS: Electrophysiology