BAE 590 SAS ML Lesson 4 Questions and Quiz
True or False: For neural networks, model generalization depends more on the number of weights than on the magnitude of the weights.
False
True or False: Like decision trees, neural networks can select inputs
False
True or False: Optimization methods are used to efficiently search the complex landscape of the error surface to find an error maximum.
False
In an MLP with one hidden layer, which of the following is a mathematical transformation that is applied to a linear combination of the input values? a. activation function b. error function c. hidden layer dhidden unit
a. activation function
Neural networks are universal approximators. This means that neural networks can do which of the following? a. model any input-output relationship, no matter how complex b. outperform any other type of model, no matter how strong the signal c. clearly indicate how the inputs affect the prediction, no matter the number of inputs
a. model any input-output relationship, no matter how complex
When early stopping is used, where does the optimal model come from? a. one of the iterations that occur earlier than the final training iteration b. the iteration immediately before the final training iteration c. the final training iteration
a. one of the iterations that occur earlier than the final training iteration
Which of the following terms refers to the intercept estimate in a neural network? a. activation function b. bias estimate c. weight estimate d. hidden unit
b. bias estimate
Which of the following techniques do deep learning models use to overcome the computational challenges associated with multiple hidden layers? a. direct connections between the input layer and the target layer b. fast moving gradient-based optimizations c. a first hidden layer in which the number of hidden units is twice the number of inputs in the input layer d. standardized inputs
b. fast moving gradient-based optimizations (i.g. SGD)
Which of the following statements is true about neural networks? a. Neural networks require that a specified form be stated prior to modeling. b. Neural networks perform best when the signal to noise ratio in a data set is low. c. Due in part to their lack of interpretability, neural networks are most relevant to pure prediction scenarios. d. Neural networks are universal approximators, so they are universally better than other types of models.
c. Due in part to their lack of interpretability, neural networks are most relevant to pure prediction scenarios.
Which of the following hyperparameters or set of hyperparameters in Model Studio controls weight decay? a. annealing rate b. learning rate c. L1 and L2 d. momentum
c. L1 and L2
Which architecture of a neural network is best for modeling data with discontinuous input-output mappings? a. multilayer perceptron with no hidden layers b. multilayer perceptron with one hidden layer c. multilayer perceptron with two hidden layers d. skip-layer perceptron
c. multilayer perceptron with two hidden layers
Which of the following occurs during a neural network's learning process? a. avoiding global minima b. late stopping c. numerical optimization
c. numerical optimization
When early stopping is used to build a neural network model, which data partition does Model Studio use to select the final model? a. test b. train c. validate
c. validate
When you are building a neural network model, which of the following methods helps to avoid overfitting? a. deviance estimation b. input standardization c. weight decay
c. weight decay
Which of the following terms refers to a parameter estimate or slope that is associated with an input in a neural network? a. activation function b. bias estimate c. weight estimate d. hidden unit
c. weight estimate
Which activation function is commonly used in the target layer when modeling a binary target? a. exponential b. hyperbolic tangent (Tanh) c. identity d. logistic
d. logistic
Which neural network architecture is best for modeling nonstationary data? a. multilayer perceptron with no hidden layers b. multilayer perceptron with one hidden layer c. multilayer perceptron with two hidden layers d. skip-layer perceptron
d. skip-layer perceptron