Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster D in Poster Session D: Wednesday, August 5, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

Hierarchical Representational Transformations of Working Memory in Brains and Machines

Qingqing Yang¹, Hyewon Willow Han², Bogeng Song³, Julie Golomb¹, Dobromir Rahnev³, Yalda Mohsenzadeh², Hsin-Hung Li¹; ¹Ohio State University, ²University of Western Ontario, ³Georgia Institute of Technology

Presenter: Qingqing Yang

Working memory (WM) maintains past inputs while processing new ones, yet how representations transform between perceptual input and retrieved memory remains unclear. Here, we analyzed human 7T fMRI data from the Natural Scenes Dataset and activations from artificial neural networks (ANNs) performing a 1-back task. Using decoding, representational similarity, and subspace geometry analyses, we quantified the general, rotational, and non-rotational changes of representations between WM encoding and retrieval phases. We found a hierarchical trend of WM coding strategy in human brains: early and intermediate visual regions exhibited larger representational changes, whereas higher-order regions maintained more stable representations. ANNs optimized to perform the same task showed a similar gradient trend, with WM stability increases along model layers, independent of model architectures and visual encoder training objective. These results established hierarchical shifts between flexibility and stability in WM representations for both biological and artificial systems.

Topic Area: Memory, Learning & Knowledge Structures