Recurrent neural network modeling of working memory

The temporary maintenance of information is known as working memory, and it enables behaviour that we take for granted, such as remembering a phone number or avoiding an obstacle after turning off the lights. Despite its importance for everyday behaviour, the neural mechanisms underlying working memory are/remain unclear. While some single-neuron recordings in monkeys correlate with working memory performance in humans, it has proven difficult to connect these sets of data with existing network models for the human brain that/as they include thousands of neurons. Here, we propose leveraging recent advances in machine learning to re-analyze existing data and bridge this gap. We will develop a flexible cortical network model by training an artificial recurrent neural network (RNN) to perform a memory task originally completed by monkeys. Individual units of the RNN will be validated with single-neuron data, and the population-level dynamics of the RNN will be compared to mechanisms proposed by existing network models.

Faculty Supervisor:

Gunnar Blohm;Martin Paré

Student:

Partner:

Philipps-Universität Marburg

Discipline:

Life Sciences

Sector:

Life Sciences (not health); Technology

University:

Queen's University

Program: