A UNIVERSAL THEOREM ON LEARNING-CURVES
A UNIVERSAL THEOREM ON LEARNING-CURVES
复制标题
DOI：
10.1016/0893-6080(93)90013-m
复制
发表时间：
1993-01-01
期刊：
影响因子：
7.8
通讯作者：
AMARI, SI
中科院分区：
文献类型：
--
作者：
AMARI, SI
学习曲线显示了随着训练样本数量的增加，学习机器改进其行为的速度有多快。本文证明了一般无噪声二分机或神经网络学习曲线的普遍渐近行为。证明了无论机器的结构如何，随着训练样本数t的增加，平均预测熵或信息增益[e*(T)]收敛到0，[e*(T)]约为d/t，其中d是机器可修改参数的个数。
A learning curve shows how fast a learning machine improves its behavior as the number of training examples increases. This paper proves a universal asymptotic behavior of learning curves for general noiseless dichotomy machines, or neural networks. It is proved that irrespective of the architecture of a machine, the average predictive entropy or the information gain [e*(t)] converges to 0 as [e*(t)] approximately d/t as the number t of training examples increases, where d is the number of modifiable parameters of a machine.