An output layer neural network node receives input from 3 hidden layer nodes and a bias. All the weights feeding into the hidden layer node are negative. During backpropagation, the differential of the output layer node's activation function with respect to the net function is deduced to be exactly 0. Which of the following are possible activation functions that the output layer node might have? Select all that apply. Negative marking applies to this question.

A dataset is divided into training, validation and testing sets. If a neural network is found to overfit, which of the following models should be selected to achieve the least effect of the overfitting in future predictions?

Which of the following statements most accurately represents the initialization of the dummy variable that represents a constant bias value? This dummy variable facilitates the dot product between a weight matrix row and an input vector as mentioned in the course material?

A binary classification model has two output nodes and uses the softmax activation function and a cross-entropy loss function. The model has derivatives of softmax and cross entropy with respect to the net function of -0.5 and 0.5, respectively. What is the net function output of the node whose ground truth is false minus the net function output of the node whose ground truth is true?

In a neural network that performs multivariable linear regression, the output layer consists of at least neuron/s.

Assume that logit1 and logit2 are logits from a neural network with two output neurons, and sm(logit1) and sm(logit2) are the corresponding softmax values for logit1 and logit2, respectively. True or false? logit1/logit2 = sm(logit1)/sm(logit2).