Neural signatures of visual memorability revealed by EEG decoding
Poster Presentation 36.320: Sunday, May 17, 2026, 2:45 – 6:45 pm, Banyan Breezeway
Session: Visual Memory: Encoding and retrieval, capacity
Schedule of Events | Search Abstracts | Symposia | Talk Sessions | Poster Sessions
Igor Utochkin1, Woohyeuk Chang1, Edward Vogel1, Edward Awh1; 1University of Chicago
Stimulus memorability is a systematic tendency for some visual items to be remembered better than others despite variations in learning and test context. Prior studies show that memorability can be decoded from neural activity. However, the extant studies have treated memorability as a single dimension. We highlight that memorability is a compound phenomenon based on two parameters: “Hittability” (H-ability), the likelihood that a studied item will be correctly recognized, and “false-alarmability” (FA-ability), the likelihood that the same item will be erroneously recognized when not studied. Strikingly, these parameters are uncorrelated, and whether their neural bases are distinct or shared remains unclear. Using multivariate decoding of EEG, we show that both H-ability and FA-ability can be robustly decoded from neural activity. Importantly, these signatures are evident not only during the test phase, where retrieval success might drive decoding, but also during study. Neural patterns associated with each parameter generalize across study and test, indicating that their representational bases are not contingent on explicit memory judgments. We next asked whether decoding of H-ability and FA-ability reflects movement along a shared representational dimension. Despite their behavioral independence, cross-decoding revealed strong generalization: classifiers trained to decode H-ability reliably decoded FA-ability, and vice versa. Thus, these parameters share a common dimension in neural state space. A natural hypothesis is that this dimension reflects item familiarity. Contrary to this prediction, neural patterns associated with high H-ability aligned with those associated with low FA-ability. This inversion rules out a simple familiarity account and instead suggests that neural signatures of memorability track systematic biases in decision efficacy based on relative familiarity. Presumably, stimuli with high H-ability and low FA-ability are easier to endorse or reject based on mnemonic status, whereas those with low H-ability and high FA-ability make these decisions more difficult.
Acknowledgements: ONR N00014-22-1-2123