Researchers have long hypothesized that our ability to learn new concepts stems from the interplay between the brain’s hippocampus and the neocortex. Hippocampus captures new information and replays it while resting and sleeping. The neocortex takes new material and looks at its existing knowledge so that it can interleave, or layer, new material into similar categories that have evolved from the past.
However, there have been some questions about this process, given that it takes the brain so long to sort through the entire storehouse of information it gathers over a lifetime. This pitfall may explain why ANNs lose long-term knowledge when absorbing new data too quickly.
Traditionally, the solution used in deep machine learning has been to retrain the network on the entire past dataset, whether it is closely related to new information or not, a very time consuming process. . UCI scientists decided to take a closer look at the problem and came up with a remarkable finding.
Advertisement
“We found that when the ANN interleaved a much smaller subset of stale information, consisting mainly of items similar to ants,” said graduate student Rajat Saxena, first author of the paper. new knowledge they acquire, they learn it without forgetting what they already know. . Saxena led the project with the support of Justin Shobe, an assistant project scientist. Both members of the laboratory of Bruce McNaughton, Distinguished Professor of Neurobiology and Behavior.
“It allows ANNs to take in new information very efficiently, without having to rethink everything they already have,” says Saxena. “These findings suggest a brain mechanism why experts in a certain field can learn new things in that field much faster than non-experts. It is only necessary in the network section. of the brain encodes expert knowledge.”
According to McNaughton, the discovery has the potential to solve cognitive problems. “Understanding the mechanisms behind learning is essential to making progress,” he said. “It gives us insights into what’s happening when the brain isn’t working the way it’s supposed to. We can develop training strategies for people with memory problems.” due to aging or people with brain damage. It could also lead to the ability to manipulate brain circuits so that people can overcome these deficits.”
Source: Eurekalert
