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

2025-02-03

AI Summary

Addressing the stability-plasticity dilemma in continual learning, this paper proposes a biologically inspired computational module—termed the Fly Model—motivated by the Drosophila olfactory circuit. The module implements sparse coding, random projection, and hierarchical nonlinear transformations in a minimalist, plug-and-play design, fully integrated end-to-end in PyTorch without introducing additional parameters. Its core contribution lies in the first generalization of fruit fly olfactory architecture principles into a low-overhead, high-generalization continual learning mechanism that simultaneously preserves stability on previously learned tasks and enhances plasticity for new ones. Evaluated on standard benchmarks—including Split-CIFAR10, Split-CIFAR100, and PermutedMNIST—the Fly Model consistently outperforms mainstream methods such as EWC and iCaRL. Crucially, it achieves near-zero parameter growth and negligible inference overhead, demonstrating strong scalability and efficiency for lifelong learning scenarios.

Technology Category

• Machine Learning: Life-Long and Continual Learning
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Artificial neural networks face the stability-plasticity dilemma in continual learning, while the brain can maintain memories and remain adaptable. However, the biological strategies for continual learning and their potential to inspire learning algorithms in neural networks are poorly understood. This study presents a minimal model of the fly olfactory circuit to investigate the biological strategies that support continual odor learning. We introduce the fly olfactory circuit as a plug-and-play component, termed the Fly Model, which can integrate with modern machine learning methods to address this dilemma. Our findings demonstrate that the Fly Model enhances both memory stability and learning plasticity, overcoming the limitations of current continual learning strategies. We validated its effectiveness across various challenging continual learning scenarios using commonly used datasets. The fly olfactory system serves as an elegant biological circuit for lifelong learning, offering a module that enhances continual learning with minimal additional computational cost for machine learning.

Problem

• Continual Learning
• Stability
• Flexibility

Innovation

• Olfactory System Inspiration
• Continual Learning
• Efficiency and Flexibility