Poster Presentation

Poster C in Poster Session C: Wednesday, August 5, 9:30 – 11:15 am, Kimmel Center, Shorin & Rosenthal Rooms

Familiarity Affects Sequence Working Memory Codes in Macaque Prefrontal Cortex and Recurrent Neural Networks

Yichen Qian¹, Wei Lun Seow¹, Camilo Libedinsky¹; ¹National University of Singapore

Presenter: Yichen Qian

Sequence working memory (SWM) is affected by familiarity of the memoranda, but the neural mechanisms underlying this familiarity remain unclear. We investigated this question by recording prefrontal cortex (PFC) activity in macaques using a 2-item SWM task. We compared neural activity associated with “overtrained” versus random or “non-overtrained” sequences. Recurrent neural networks (RNNs) were trained to perform the same task to investigate computational mechanisms. Behavioral results showed a more accurate recollection of the overtrained sequences in both animals and RNNs. Neural decoding analyzes indicated that both individual item and sequence identity information were decodable by the activity of the PFC population, although this information was found primarily during the late delay period instead of being persistent as previously reported. Neural decoders distinguished the overtrained sequences more accurately and obtained biased predictions favoring the overtrained second items (I2) given corresponding first items (I1). We further investigated the low-dimensional representation of sequential items and whether the animal memorized the sequence as one integrated object or as two separate objects. We first estimated a one-dimensional representational space for each item, and we found that I1 and I2 spaces were near-orthogonal in macaque PFC. We then constructed a 2D space representing I1-I2 sequences by spanning the two 1D spaces. Projections showed that I1-I2 identity was coded as the composition of I1 and I2. We further asked whether overtraining affected the sequences representational configurations on this space, and we found the overtrained sequences were more differentiated in the space. Altogether, our results revealed that familiarity warps the representational space of WM in prefrontal cortex.

Topic Area: Memory, Learning & Knowledge Structures