AI Engram: In Search of Memory Traces in Artificial Intelligence

AI Engram：探寻人工智能中的记忆痕迹

Abstract: Memory formation is fundamental to intelligence, yet whether deep neural networks preserve identifiable memory traces analogous to biological memory units remains an open question.

摘要： 记忆的形成是智能的基础，然而深度神经网络是否保留了类似于生物记忆单元的可识别记忆痕迹，仍是一个悬而未决的问题。

This work introduces a geometric framework to identify such “AI engrams” by formalizing the neuroscientific criteria of specificity, reactivation, sufficiency, and necessity into a constrained inverse problem.

本研究引入了一个几何框架来识别此类“AI记忆痕迹（AI engrams）”，通过将神经科学中的特异性、重激活、充分性和必要性标准形式化为一个受约束的逆问题。

We derive a closed-form estimator that isolates individual memory traces from globally entangled parameters, and show that this biologically-derived solution corresponds to a natural gradient update on the parameter manifold.

我们推导出一个闭式估计器，能够从全局纠缠的参数中分离出独立的记忆痕迹，并证明这种源于生物学的解决方案对应于参数流形上的自然梯度更新。

AI engrams enable surgical manipulation of learned knowledge: any subset of memories can be composed or erased through linear arithmetic, without iterative optimization.

AI记忆痕迹实现了对所学知识的“外科手术式”操作：无需迭代优化，仅通过线性算术即可组合或擦除任何记忆子集。

Experiments ranging from simple MLPs to LLMs demonstrate the causal validity and substantial scalability of AI engrams.

从简单的多层感知机（MLP）到大语言模型（LLM）的实验证明了AI记忆痕迹的因果有效性和强大的可扩展性。

Together, these results bridge theories of biological memory and artificial representation learning and offer geometric insight into how deep networks simultaneously support functional specificity within distributed storage.

总之，这些研究结果架起了生物记忆理论与人工表征学习之间的桥梁，并为深度网络如何在分布式存储中同时支持功能特异性提供了几何学视角。