Physics Magazine recently published a web article authored by Assistant Professor of Physics Sarah Marzen, “Time Delays Improve Performance of Certain Neural Networks.” In the piece, Marzen describes how reservoir computers function as artificial neural networks and can predict future system behavior based on past information. However, the predictive abilities of these computers are imperfect due to challenges with information storage and loss, which also makes the computers’ construction resource intensive.
New research adding time delays from S. Kamyar Tavakoli and André Longtin of the University of Ottawa, Canada, points to improving both the reservoir computer’s memory capacity and its predictive capabilities. Marzen, who has also conducted research on reservoir computers, says Tavakoli and Longtin’s outcomes signal progress in creating an optimal artificial neural network and prompt more questions about scientists’ functional understanding of the brain in biological organisms.