Structural features of the fly olfactory circuit mitigate the stability-plasticity dilemma in continual learning

  • 2025-02-03 15:06:11
  • Heming Zou, Yunliang Zang, Xiangyang Ji
  • 0

Abstract

Artificial neural networks face the stability-plasticity dilemma in continuallearning, while the brain can maintain memories and remain adaptable. However,the biological strategies for continual learning and their potential to inspirelearning algorithms in neural networks are poorly understood. This studypresents a minimal model of the fly olfactory circuit to investigate thebiological strategies that support continual odor learning. We introduce thefly olfactory circuit as a plug-and-play component, termed the Fly Model, whichcan integrate with modern machine learning methods to address this dilemma. Ourfindings demonstrate that the Fly Model enhances both memory stability andlearning plasticity, overcoming the limitations of current continual learningstrategies. We validated its effectiveness across various challenging continuallearning scenarios using commonly used datasets. The fly olfactory systemserves as an elegant biological circuit for lifelong learning, offering amodule that enhances continual learning with minimal additional computationalcost for machine learning.

 

Quick Read (beta)

loading the full paper ...