Notes

• Centralized AI faces scaling bottlenecks (data transmission, latency, energy, data center reliance) at the edge.
• Node Learning places intelligence at individual edge nodes, expanding knowledge via selective peer interaction.
• Nodes learn continuously from local data, maintain own model state, and exchange knowledge opportunistically.
• Learning propagates via overlap and diffusion, not global synchronization or central aggregation.
• The paradigm unifies autonomous/cooperative behavior and accommodates heterogeneity in data, hardware, objectives, and connectivity.
• The paper examines implications for communication, hardware, trust, and governance, offering a broader decentralized perspective.

• 集中式AI在边缘场景面临数据传输、延迟、能耗和数据中心依赖等扩展瓶颈。
• Node Learning 将智能置于单个边缘节点，通过选择性对等交互扩散知识。
• 节点持续从本地数据学习，维护自身模型状态，并在有益时进行机会性知识交换。
• 学习通过重叠和扩散传播，而非全局同步或中心化聚合。
• 该范式在单一抽象中统一了自主和协作行为，适应数据、硬件、目标和连接性的异构性。
• 论文探讨了该范式对通信、硬件、信任和治理的影响，为现有去中心化方法提供了更广阔的视角。