题目
多项选择题
Question25 You are training a neural network but your training error is very high. Which of the following (if done in isolation) approaches might help? (You can choose more than one):Select one or more alternatives: Normalizing the input data Adding more training data Adding more units to each hidden layer Adding another hidden layer ResetMaximum marks: 1.5 Flag question undefined
选项
A.Normalizing the input data
B.Adding more training data
C.Adding more units to each hidden layer
D.Adding another hidden layer
查看解析
标准答案
Please login to view
思路分析
When diagnosing why training error is very high, we need to consider strategies that directly affect the model's ability to fit the training data as well as the learning process itself.
Option 1: Normalizing the input data. This often helps because features on different scales can cause the gradient updates to be uneven, leading to slower convergence or getting stuck in suboptimal regions. By standardizing or normalizing inputs, the optimization landscape becomes more stable, which can reduce training error more quickly. This is a well-known......Login to view full explanation登录即可查看完整答案
我们收录了全球超50000道考试原题与详细解析,现在登录,立即获得答案。
类似问题
Which of the following statements about Neural Network (NN) is correct?
Which of the following correctly describes a feedforward neural network? I All of its units, called neurons, in a hidden layer are connected to all of the units in a previous layer. II Its goal is to approximate a function 𝑓 ( 𝑋 ) of the input 𝑋 to get an output 𝑦 . III Feedforward refers to the direction of the information flows through the function being evaluated from 𝑋 up to the output 𝑦 . IV They have feedback connections where the output of a certain layer is fed back to itself, making them the best choice for modeling time series.
What is true regarding the neural network family of algorithms? I Neural Networks can have at most two hidden layers. II Neural Network training involves the computation of several partial derivatives of the error function concerning the model parameters. III Neural Networks can be used as a supervised and an unsupervised method. IV Neural Networks can be used only as a supervised method.
What is true regarding an Autoencoder (AE)? I It is a neural network with a specific structure. II It is an unsupervised learning algorithm. III It is an algorithm to train a regression tree. IV It is a supervised learning algorithm.
更多留学生实用工具
希望你的学习变得更简单
加入我们，立即解锁 海量真题 与 独家解析，让复习快人一步！