Search Papers | Poster Sessions | All Posters
Poster A39 in Poster Session A - Tuesday, August 6, 2024, 4:15 – 6:15 pm, Johnson Ice Rink
Fuzzy learning can decouple language and odor representations: Associative memory loss in a computational model
Nikolaos Chrysanthidis1 (), Rohan Raj2, Thomas Hörberg2, Robert Lindroos2, Anders Lansner1,3, Jonas Olofsson2, Pawel Herman1,4; 1KTH Royal Institute of Technology, Department of Computational Science and Technology, Stockholm, Sweden, 2Stockholm University, Department of Psychology, Stockholm, Sweden, 3Stockholm University, Department of Mathematics, Stockholm, Sweden, 4Digital Futures, KTH Royal Institute of Technology, Stockholm, Sweden
Odor naming is considered as a particularly challenging test of olfactory ability. It is common that in free naming scenarios people fail to respond with any linguistic label to certain odors they smell, resulting in an omission (i.e., a blank response in a test). The cognitive demands and the nature of interactions between olfactory memory and language related brain’s neural resources are not well understood. Here with the support of a computational model we offer some hypotheses regarding the neural network-level mechanisms underlying the phenomenon of omissions in a free odor naming task. The available behavioral data suggests that odors with numerous language associations (one-to-many mapping) lead to elevated blank responses. To simulate the task and examine whether the trends observed in the behavioral data can be reproduced and mechanistically analyzed, we developed a memory model consisting of two networks that are reciprocally connected with Bayesian-Hebbian plasticity [1]. We stored and associated distributed, overlapping odor percept and odor label language representations in this model network. Overall, we suggested and evaluated a hypothesis that associative Bayesian-Hebbian plasticity for the connections between the two networks results in weak coupling for odors paired with multiple different labels during the odor-label encoding (one-to-many mapping), thereby increasing the subthreshold network responses (omissions) for these odors.
Keywords: Bayesian-Hebbian Plasticity Olfactory Memory Odor-naming task Neural network model