是否存在使用感知器算法的情况，我不想在每次迭代之前排列数据? [英] Are there cases using the Perceptron algorithm where I do NOT want to permute the data before every iteration?

问题描述

我的机器学习教科书问了这个问题，讨论了感知器算法，但我真的无法给出满意的答案.

My machine learning textbook asks this question, discussing the perceptron algorithm, and I really can't come up with a satisfied answer.

有什么情况?

推荐答案

与任何没有全局最优解的迭代学习算法一样，感知器算法将从起点收敛到局部最优解.通常，这意味着早期数据的影响将大于晚期数据.

Like any iterative learning algorithm with no globally optimal solution the perceptron algorithm will converge from a starting point to a locally optimal solution. This usually means the early data will have a larger influence than the later.

在感知器算法的大多数应用中，您尝试通过以随机顺序多次应用训练数据来消除这种偏差.

In most applications of a perceptron algorithm you try to eliminate this bias by multiple application of the training data in random order.

在某些应用中，此偏差是学习问题的一部分，因此顺序很重要，并且最终结果更好，没有随机化.

In some application, this bias is part of the learning problem, so the order matters and the final result is better with no randomization.

这篇关于是否存在使用感知器算法的情况，我不想在每次迭代之前排列数据?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！