Dynamics of learning
Dynamics of learning
复制标题
学习动力
DOI：
10.1007/978-3-642-79814-6_4
复制
发表时间：
1991
期刊：
影响因子：
--
通讯作者：
M. Opper
中科院分区：
文献类型：
--
作者：
W. Kinzel;M. Opper
研究了具有联想记忆功能的吸引子网络中的有监督学习。从最优化问题出发，给出了两种学习算法：最优稳定性的PERCEPTRON算法和Adaline算法，并得到了它们的精确动态结果。将Adaline推广到具有二元突触的网络，并对其进行了数值研究。学习过程中的吸引力盆地是使用高斯近似计算的。最后给出了在Adaline网络中遗忘的分析结果。
Supervised leaming in attractor networks which perform as an associative memory is investigated. Two leaming algorithms, the PERCEPTRON of optimal stability and the ADALINE, are derived from optimization problems and exact results for their dynamics are obtained. The ADALINE is extended to networks with binary synapses and is studied numerically. The basins of attraction during the leaming process are calculated using a Gaussian approximation. Finally analytical results for forgetting in the ADALINE network are presented.