Forgetful neural networks could shed light on how humans learn
How the brain learns is one of the biggest unsolved mysteries in science. Researchers in artificial intelligence from the University of Groningen (the Netherlands) and KU Leuven (Belgium) hope that a specific type of forgetting in artificial neural networks will help to capture what is happening in the human brain. The study is published in the journal Trends in Cognitive Sciences.
The mystery of how the brain learns remains one of science's greatest enigmas. Artificial neural networks, models inspired by the human brain, offer a potential window into this process. These networks learn tasks through a method called gradient descent, where they incrementally adjust the connections between units to optimize performance.
However, once trained on a particular task (like recognizing a boat), the network often struggles to maintain that knowledge when learning a new task (like recognizing a car). This phenomenon, known as catastrophic forgetting, can be mitigated through a technique called replay, where the network is periodically reminded of the earlier task during the learning of the new one.
While replay has proven effective in many cases, researchers have recently discovered a peculiar pattern in the network's performance—a transient forgetting known as the stability gap. This gap refers to the temporary decline in the network's ability to perform the old task while learning the new one, even though the network continues to practice the old task.
Intriguingly, scientists from the University of Groningen and KU Leuven hypothesized that a similar stability gap might occur in human learning. They conducted a review of existing cognitive science literature and found multiple instances of transient forgetting in humans, but none that directly mirrored the stability gap observed in neural networks.
Crucially, the absence of evidence against a stability gap in human learning also leaves the possibility open. The researchers argue that detecting a stability gap in human learning would provide valuable insights into the mechanisms of human cognition. One key question is whether the human brain learns through a process akin to gradient descent.
If such a stability gap is found in future human learning studies, it would strongly suggest that our brains employ a learning process similar to that of artificial neural networks. Conversely, if the stability gap is not present in humans, it would imply that the human brain utilizes a fundamentally different learning mechanism.
Written by urgent.news from Medical Xpress's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.