AI Models Mimic Human Memory Consolidation, Offering Paths to Improved AI and Insights into the Brain
-
Researchers discovered AI models like Transformer mimic human memory consolidation processes, specifically in the hippocampus, offering potential AI advancements and deeper understanding of human memory.
-
The team applied principles of memory consolidation in the brain's NMDA receptor to the Transformer model to try improving its memory function.
-
Just as changing magnesium levels affects memory strength in the brain, tweaking parameters in the Transformer to reflect the NMDA receptor's gating action improved the model's memory.
-
This finding suggests established neuroscience knowledge can explain AI learning mechanisms.
-
Converging human cognitive principles and AI design shows promise for better AI systems and for gleaning insights into the brain through AI models.