(associative memory / parallel processing / categorization / content-addressable memory / fail-soft devices) Contributed by John J. Hopfield, January 15, 1982 ABSTRACT Computational properties of use to biological organisms or to the construction of computers can emerge as collective
核心贡献 · Key contributions
提出神经网络作为物理系统具有涌现的集体计算能力。 Proposes that neural networks as physical systems exhibit emergent collective computational abilities.
利用异步并行处理和对称连接实现内容寻址记忆。 Implements a content-addressable memory using asynchronous parallel processing and symmetric connections.
证明了能量函数单调递减，确保收敛到稳定状态。 Demonstrates an energy function that monotonically decreases, ensuring convergence to stable states.
展示了存储约 0.15N 个记忆的能力，具有纠错和模式补全功能。 Shows capacity to store approximately 0.15N memories with error correction and pattern completion.
揭示了涌现特性：泛化、熟悉度识别、分类和时间序列保持。 Reveals emergent properties: generalization, familiarity recognition, categorization, and time sequence retention.
表明鲁棒的容错性和适用于硬件实现。 Suggests robust fault tolerance and suitability for hardware implementation.
局限 · Limitations
假设对称连接（Tij = Tji），在生物学上不现实。 Assumes symmetric connections (Tij = Tji), which is biologically unrealistic.
记忆容量限制在约 0.15N，随网络规模扩展性差。 Memory capacity limited to about 0.15N, scaling poorly with network size.
要求存储的记忆近乎随机且不相关以实现最优性能。 Requires stored memories to be nearly random and uncorrelated for optimal performance.
无法可靠编码超过几个状态的时间顺序序列。 Unable to reliably encode long sequences of temporal order beyond a few states.
模型省略了许多神经生物学细节，如突触延迟和随机脉冲发放。 Model omits many neurobiological details such as synaptic delays and stochastic spiking.
论文章节 · Sections（共 10）
计算能力computational abilities
随时间进行直到 X Xa。我们可以将信息视为proceed in time until X Xa. We can regard the information
V. 引言Vi0if IT.,V. joi
伪正交性Tijjs =E (2V, - 1) I VJ(2Vj-1) Hjs. 3
生物物理学：Hopfield 网络Biophysics: Hopfield Downloaded from https://www.pnas.org by 34.96.52.10 on July 29, 2026 from IP address 34.96.52.10. Proc. NatL Acad. Sci. USA 79 (1982) becomes an input-output relationship for a neuron.
Tjj？如果在某个时刻算法将 Vi 从 0 变为 1 或反之，则公式(7)定义的能量变化可以Tjj? If the algorithm at some time changes Vi from 0 to 1 or vice versa, the change of the energy defined in Eq. 7 can be
Tt 可在此范围内自由递增Tt, be freely incremented within this range. If Tij = 3, a next
生物物理学：霍普菲尔德网络Biophysics: Hopfield Downloaded from https://www.pnas.org by 34.96.52.10 on July 29, 2026 from IP address 34.96.52.10. Proc. Natl. Acad. Sci. USA 79 (1982)
基于部分神经元的存储容量分析ATUj = (2Xi -1)(2Xj -1) Q~ -- k < N ill
ATU = A > (2^{V_s+1} - 1)(2^{V_j} - 1) 13ATU = A > (2Vs+1 - 1)(2Vj - 1) 13