通过选择进行学习的自旋玻璃模型。
Spin glass model of learning by selection.
作者信息
Toulouse G, Dehaene S, Changeux J P
出版信息
Proc Natl Acad Sci U S A. 1986 Mar;83(6):1695-8. doi: 10.1073/pnas.83.6.1695.
Abstract
A model of learning by selection is described at the level of neuronal networks. It is formally related to statistical mechanics with the aim to describe memory storage during development and in the adult. Networks with symmetric interactions have been shown to function as content-addressable memories, but the present approach differs from previous instructive models. Four biologically relevant aspects are treated--initial state before learning, synaptic sign changes, hierarchical categorization of stored patterns, and synaptic learning rule. Several of the hypotheses are tested numerically. Starting from the limit case of random connections (spin glass), selection is viewed as pruning of a complex tree of states generated with maximal parsimony of genetic information.
摘要
在神经网络层面描述了一种通过选择进行学习的模型。它在形式上与统计力学相关，旨在描述发育过程中和成年期的记忆存储。具有对称相互作用的网络已被证明可作为内容可寻址存储器，但当前方法与先前的指导性模型不同。探讨了四个与生物学相关的方面——学习前的初始状态、突触符号变化、存储模式的层次分类以及突触学习规则。对其中几个假设进行了数值测试。从随机连接（自旋玻璃）的极限情况出发，选择被视为对以最大简约遗传信息生成的复杂状态树的修剪。
相似文献
1
Spin glass model of learning by selection.通过选择进行学习的自旋玻璃模型。
Proc Natl Acad Sci U S A. 1986 Mar;83(6):1695-8. doi: 10.1073/pnas.83.6.1695.
2
Learning from mistakes.从错误中学习。
Neuroscience. 1999;90(4):1137-48. doi: 10.1016/s0306-4522(98)00472-2.
3
Storage capacity diverges with synaptic efficiency in an associative memory model with synaptic delay and pruning.在具有突触延迟和修剪的联想记忆模型中，存储容量与突触效率不同。
IEEE Trans Neural Netw. 2004 Sep;15(5):1215-27. doi: 10.1109/TNN.2004.832711.
4
Reinforcement learning with modulated spike timing dependent synaptic plasticity.基于调制的尖峰时间依赖突触可塑性的强化学习。
J Neurophysiol. 2007 Dec;98(6):3648-65. doi: 10.1152/jn.00364.2007. Epub 2007 Oct 10.
5
Neural associative memory with optimal Bayesian learning.最优贝叶斯学习的神经联想记忆。
Neural Comput. 2011 Jun;23(6):1393-451. doi: 10.1162/NECO_a_00127. Epub 2011 Mar 11.
6
Formation and Maintenance of Robust Long-Term Information Storage in the Presence of Synaptic Turnover.在突触更替情况下稳健长期信息存储的形成与维持
PLoS Comput Biol. 2015 Dec 29;11(12):e1004684. doi: 10.1371/journal.pcbi.1004684. eCollection 2015 Dec.
7
"Unlearning" increases the storage capacity of content addressable memories.“忘却”可增加内容可寻址存储器的存储容量。
Biophys J. 1987 Jan;51(1):47-53. doi: 10.1016/S0006-3495(87)83310-6.
8
Statistical mechanics of reward-modulated learning in decision-making networks.决策网络中受奖励调节的学习的统计力学。
Neural Comput. 2012 May;24(5):1230-70. doi: 10.1162/NECO_a_00264. Epub 2012 Feb 1.
9
Learning only when necessary: better memories of correlated patterns in networks with bounded synapses.仅在必要时学习：具有有限突触的网络中对相关模式的更好记忆。
Neural Comput. 2005 Oct;17(10):2106-38. doi: 10.1162/0899766054615644.
10
Reinforcement learning, spike-time-dependent plasticity, and the BCM rule.强化学习、尖峰时间依赖性可塑性与BCM规则。
Neural Comput. 2007 Aug;19(8):2245-79. doi: 10.1162/neco.2007.19.8.2245.
引用本文的文献
1
Lattice physics approaches for neural networks.神经网络的晶格物理方法。
iScience. 2024 Nov 15;27(12):111390. doi: 10.1016/j.isci.2024.111390. eCollection 2024 Dec 20.
2
Neural Activity in Quarks Language: Lattice Field Theory for a Network of Real Neurons.夸克语言中的神经活动：真实神经元网络的格点场论
Entropy (Basel). 2024 Jun 6;26(6):495. doi: 10.3390/e26060495.
3
Spatial evolution of the proton-coupled Mott transition in correlated oxides for neuromorphic computing.用于神经形态计算的关联氧化物中质子耦合莫特转变的空间演化。
Sci Adv. 2024 May 31;10(22):eadk9928. doi: 10.1126/sciadv.adk9928.
4
Ordering Dynamics in Neuron Activity Pattern Model: An Insight to Brain Functionality.神经元活动模式模型中的排序动力学：对脑功能的洞察
PLoS One. 2015 Oct 27;10(10):e0141463. doi: 10.1371/journal.pone.0141463. eCollection 2015.
5
Artificial neural network modeling for deciding if extractions are necessary prior to orthodontic treatment.人工神经网络模型用于决定正畸治疗前是否需要拔牙。
Angle Orthod. 2010 Mar;80(2):262-6. doi: 10.2319/111608-588.1.
6
Hebbian learning reconsidered: representation of static and dynamic objects in associative neural nets.重新审视赫布学习：关联神经网络中静态和动态对象的表征
Biol Cybern. 1989;60(6):457-67. doi: 10.1007/BF00204701.
7
An n-level field theory of biological neural networks.生物神经网络的n层场论。
J Math Biol. 1993;31(8):771-95. doi: 10.1007/BF00168045.
8
Exact results for the average dynamic behavior of some non-linear neural networks.一些非线性神经网络平均动态行为的精确结果。
Biol Cybern. 1987;57(4-5):249-56. doi: 10.1007/BF00338818.
9
A cognitive and associative memory.一种认知和联想记忆。
Biol Cybern. 1987;57(3):197-206. doi: 10.1007/BF00364151.
10
Neural networks that learn temporal sequences by selection.通过选择来学习时间序列的神经网络。
Proc Natl Acad Sci U S A. 1987 May;84(9):2727-31. doi: 10.1073/pnas.84.9.2727.
本文引用的文献
1
Storing infinite numbers of patterns in a spin-glass model of neural networks.在神经网络的自旋玻璃模型中存储无限数量的模式。
Phys Rev Lett. 1985 Sep 30;55(14):1530-1533. doi: 10.1103/PhysRevLett.55.1530.
2
Spin-glass models of neural networks.神经网络的自旋玻璃模型。
Phys Rev A Gen Phys. 1985 Aug;32(2):1007-1018. doi: 10.1103/physreva.32.1007.
3
Neural networks and physical systems with emergent collective computational abilities.具有涌现集体计算能力的神经网络与物理系统。
Proc Natl Acad Sci U S A. 1982 Apr;79(8):2554-8. doi: 10.1073/pnas.79.8.2554.
4
'Unlearning' has a stabilizing effect in collective memories.“忘却”对集体记忆具有稳定作用。
Nature. 1983;304(5922):158-9. doi: 10.1038/304158a0.
5
Collective properties of neural networks: a statistical physics approach.神经网络的集体特性：一种统计物理学方法。
Biol Cybern. 1984;50(1):51-62. doi: 10.1007/BF00317939.
6
Regressive events in neurogenesis.神经发生中的退行性事件。
Science. 1984 Sep 21;225(4668):1258-65. doi: 10.1126/science.6474175.
7
8
[Molecular model of the regulation of chemical synapse efficiency at the postsynaptic level].[突触后水平化学突触效率调节的分子模型]
C R Seances Acad Sci III. 1982 Dec 6;295(12):665-70.
9
A model for the origin of biological catalysis.生物催化起源的一种模型。
Proc Natl Acad Sci U S A. 1984 Mar;81(6):1751-3. doi: 10.1073/pnas.81.6.1751.
10
A theory of the epigenesis of neuronal networks by selective stabilization of synapses.一种通过突触的选择性稳定化来解释神经网络表观遗传的理论。
Proc Natl Acad Sci U S A. 1973 Oct;70(10):2974-8. doi: 10.1073/pnas.70.10.2974.
Zotero 插件