Search Papers | Poster Sessions | All Posters
Poster C40 in Poster Session C - Friday, August 9, 2024, 11:15 am – 1:15 pm, Johnson Ice Rink
Investigating the Emergence of Complexity from the Dimensional Structure of Mental Representations
Trisha Mazumdar1, Vivian Cai1, Stephen Zhao1, Jason Lee1, Andrew Bender2, Adaline Leong1, Jad El Harake1, Mark Wallace1, David Tovar1 (); 1Vanderbilt University, 2University of California San Diego
Objects can be described by various dimensions that, when combined, form a distinct entity. This study explores the multi-dimensional structure of mental representations of objects and the associated property of complexity. We investigated how different characteristics influence perceived complexity and evaluated the predictive power of entropy scores as indicators of this complexity. Our results show that entropy scores, calculated from mental embeddings and adjusted by perceptual weights, can predict perceptual complexity effectively. Notably, once these weights are tuned to the relative complexity of each dimension, entropy scores based on human complexity ratings significantly enhance the correlation between entropy and participant choices in distinguishing between ambiguous and control images. Importantly, we established a complexity score using a perceptually tuned CLIP model, CLIP-HBA, that makes this metric generalizable to novel stimuli due to its ability to detect perceptually relevant dimensions in objects.
Keywords: object recognition complexity mental embeddings ambiguity