Representation of memories in an abstract synaptic space and its evolution with and without sleep. Credit: Golden R, Delanois JE, Sanda P and Bazhenov M, 2022, PLOS Computational Biology, CC-BY 4.0 (creativecommons.org/licenses/by/4.0/)
A trio of researchers from the University of California, working with a colleague from the Institute of Computing at the Czech Academy of Sciences, have found that it is possible to prevent catastrophic forgetting in AI systems by causing these systems to mimic human REM sleep.
In their article published in Computational Biology PLOSRyan Golden, Jean Erik Delanois, Maxim Bazhenov and Pavel Sanda describe teaching artificial intelligence systems remember what was learned from an initial task when working on a second task.
Previous research has shown that people experience what is called consolidation of Memory during REM sleep. It is a process by which things that have been experienced recently are moved to long term memory Make room for new experiences. Without such a process, the brain experiences catastrophic forgetting, where memories of recent things are not retained.
This is seen in some older people who lose the ability to sleep well, and thus find themselves able to remember things from the distant past but not things that have happened in recent days. In this new effort, the researchers found that something similar can be used to help AI systems retain what has been learned in the past while learning new things.
AI systems have become well known for their ability to master a certain genre – one can create one to become a chess master, for example. But getting AI systems to master more than one topic has proven difficult.
Indeed, explain the researchers, new learning tends to be at the expense of old ones. The more one learns something in a new area, the more old memories are lost until they disappear completely. To overcome this problem, researchers looked at how the human brain handles similar situations.
First, they built an AI system that first learned one task and then another. They found, as expected, that as he improved in the second task, he lost his ability in the first task. To overcome the problem, the researchers added code that mimicked REM sleep in the human brain. They essentially gave the system the ability to intersperse sleep/work phases which allowed the system to continue to retain older memories as new ones were processed and which helped prevent catastrophic forgetfulness.
Sleep prevents catastrophic forgetting in advanced neural networks by forming a joint representation of synaptic weight, Computational Biology PLOS (2022). journal.plos.org/ploscompbiol … journal.pcbi.1010628
© 2022 Science X Network
Quote: Imiting human sleep as a way to prevent catastrophic forgetting in AI systems (2022, November 24) retrieved November 25, 2022 from https://techxplore.com/news/2022-11-mimicking-human-catastrophic-ai.html
This document is subject to copyright. Except for fair use for purposes of private study or research, no part may be reproduced without written permission. The content is provided for information only.
