核心概念
本文提出了一種新的動態範式來模擬大腦功能,特別是記憶的運作方式,透過引入受控變量來調整神經元之間的交互作用,從而改進經典的霍普菲爾德模型。
統計資料
N ≈ 10^11 (神經元數量)
10^13 ≲ M ≲ 10^15 (活躍連接數量)
引述
"The renowned Hopfield model‡ for neural networks [8] consists of N neurons, each with an initial electric status V (0) i = 0, 1 (1: firing, 0: not firing), interacting through synaptic weights Tij, representing the strength of connection."
"Nicolas Brunel§, in a 2022 VIMM conference, emphasized the need for generalized Hopfield models where the synaptic weights Tij depend on potentials Vi, rather than being constant [6]."
"This proposal aligns with Hebbian learning, where the synaptic matrix updates as: T old ij −→T new ij = T old ij + 1 N bVi bVj."