Energetically efficient learning in neuronal networks

• Published in: Current opinion in neurobiology 2023-12, Vol.83, p.102779-102779, Article 102779
• Main Authors: Pache, Aaron, van Rossum, Mark C.W.
• Format: Article
• Language: English
• Subjects: Animals; Humans; Learning - physiology; Models, Neurological; Neural Networks, Computer; Neuronal Plasticity - physiology; Neurons - physiology
• ISSN: 0959-4388; 1873-6882
• DOI: 10.1016/j.conb.2023.102779

Human and animal experiments have shown that acquiring and storing information can require substantial amounts of metabolic energy. However, computational models of neural plasticity only seldom take this cost into account, and might thereby miss an important constraint on biological learning. This review explores various ways to reduce energy requirements for learning in neural networks. By comparing the resulting learning rules to cognitive and neurophysiological observations, we discuss how energy efficiency might have shaped biological learning. •Memory formation can be metabolically costly.•Biological learning needs to be energy efficient, but computational theories tend to ignore this constraint.•Orders of magnitude of energy can be saved by tweaking standard learning rules.•Energy efficient learning is consistent with cognitive and neurophysiological findings.