New research involving LMH academic Associate Professor Rui Ponte Costa has shed light on how the brain can learn from new experiences without interfering with what it has already learned.
Published in Nature, the study reveals how two parts of the brain (the neocortex and cerebellum) may work together to allow us to apply existing knowledge to new situations while keeping different experiences distinct.
Researchers tracked activity in the premotor cortex and cerebellum as mice learned two different tasks that shared the same basic structure: performing an action, waiting, and receiving a reward.
They discovered a division of labour between the two brain regions. The cortex reused similar patterns of activity across the related tasks, providing shared building blocks for behaviour. The cerebellum, meanwhile, reorganised these patterns to distinguish between different situations.
Discussing the findings, Associate Professor Costa said: “The fact that we find a solution to how the brain can generalise across experiences while contextualising them is remarkable.”
The findings provide new insight into a fundamental challenge faced by the brain: how do we continually learn while preventing different skills and experiences from interfering with one another?
The study was a collaboration between researchers at the University of Oxford and the US National Institutes of Health (NIH). Experimental work was led by Martha Garcia-Garcia in Mark Wagner’s laboratory at the NIH, while computational modelling at Oxford was led by Michał Wójcik, a postdoctoral researcher in the Costa Group in the Department of Physiology, Anatomy and Genetics.
The researchers used computational models alongside the experimental findings to investigate how different neural architectures can support learning while avoiding interference between tasks. Their results challenge the traditional idea that the cerebellar granule cells mainly separate information by transforming it into increasingly complex, high-dimensional representations. The study suggests the cerebellum may separate experiences while preserving what they have in common.
The work could also have implications beyond neuroscience. Artificial intelligence systems face a common challenge in learning continually and adapting to new situations without losing previously acquired knowledge.
Associate Professor Costa concluded: “This means that we now potentially understand how we can continue learning without forgetting previous experiences. Also offering a potential way forward to the same issues faced by the current inability for life-long-learning and lack of adaptability faced by AI systems.”
Read the paper here.
