Wuxing Memory: A Resource-Competition Framework for Continual Learning

This paper proposes the Wuxing Memory Architecture (WMA), a resource-competition framework for continual learning. Inspired by the mutual-generation (xiangsheng) and mutual-overcoming (xiangke) mechanisms of Eastern philosophy, WMA decomposes the memory system into five heterogeneous modules: Exploration (Wood), Reasoning (Fire), Knowledge (Earth), Rules (Metal), and Adaptation (Water). Under finite resource constraints, WMA allows dynamic balance to emerge from local interactions, effectively addressing the "catastrophic forgetting" problem in artificial intelligence. 本文提出了一种基于“五行”逻辑的内存架构(WMA)。通过引入资源竞争机制,模拟东方哲学中的“生克”原理,将内存系统划分为探索、推理、知识、规则与适配五个模块。该架构通过动态平衡解决 AI 持续学习中的资源分配与遗忘问题,实现了从反应式智能(L1)向学习式智能(L2)的跨越。