Loading…
CNS*2020 Online has ended
Welcome to the Sched instance for CNS*2020 Online! Please read the instruction document on detailed information on CNS*2020.
Back To Schedule
Sunday, July 19 • 8:00pm - 9:00pm
P203: A neural mechanism of working-memory manipulation in a categorization-task performance

Log in to save this to your schedule, view media, leave feedback and see who's attending!

Feedback form is now closed.
Yoshiki Kashimori, Hikaru Tokuhara

Working memory has a function by which temporal information is maintained and recognized in the brain, and is ubiquitous in various brain regions. A growing body of working memory research has indicated the neural mechanism underlying the maintenance of working memory. Several studies have demonstrated that working memory is maintained by a persistent activity of neural assemblies. Other studies have proposed that it is maintained by a short-term synaptic plasticity. The mechanism of working memory maintenance is still a matter of debate. Furthermore, it is unclear how working memory is linked to behavior and decision-making.

In this study, to clarify the neural mechanisms underlying the maintenance and manipulation of working memory, we focus on the function of prefrontal cortex in a delayed match-to-categorization task studied by Freedman et al. [1]. In this task, monkeys were presented with a sample and a test stimulus, separated by a delay period, and were trained to judge whether these stimuli were from the same category. Freedman et al. showed that working memory of category information was formed in the PFC. Our previous model demonstrated the neural mechanism of the working memory shaped in the PFC [2]. In this study, we aim to understand a unified mechanism of working memory maintenance and its manipulation for behavior. We develop a network model that performs the maintenance and recognition of temporal information of a sample and a test stimulus. The model consists of the networks of IT and PFC. The PFC model is further constructed with a positive-feedback-loop layer, a recurrent network, and a decision layer. The positive-feedback-loop layer produces a persistent activity of a previously presented stimulus, allowing the layer to maintain information of a sample stimulus as working memory. The recurrent network encodes the temporal information of a sample stimulus and a test stimulus. The learning of temporal information was made by Backpropagation Through Time method. The decision layer has neurons responding to a match and a non-match trial. We also investigate the discrimination ability of our model for more complex tasks that have longer temporal sequences and many category numbers.

We demonstrate that maintenance of working memory and encoding of temporal sequence are sequentially manipulated in different areas of the PFC. We also show that the temporal sequence is encoded by activity pattern of the recurrent circuit, independently of task decision. The sparseness of activity pattern increases with increasing the number of category. The principal component analysis of activity patterns reveals that the activity patterns of non-match trials move far away from the activity patterns of match trials as the learning proceeds. Furthermore, we show that the decision of task trials is adjusted by the learning of the connections between recurrent neurons representing the activity patterns and decision neurons, according to task context.

[1] Freedman, DJ et al. J Neurosci 23, 5235-5246, 2003.

[2] Abe, Y et al. Cog Comput 10, 687–702, 2018.

Speakers
YK

Yoshiki Kashimori

Dept. of Engineering Science, The University of Electro-Communications



Sunday July 19, 2020 8:00pm - 9:00pm CEST
Slot 15