Quiz 3 - Question 1

You are training the weights of a single-neuron classifier with a sigmoid activation function that predicts either the token “rice” (positive class) or the token “cake” (negative class) from a two-dimensional prompt embedding. Your classifier also uses dropout in the input layer, that is with some probability, it sets each component of the input vector to 0. The current weights of the classifier are represented by the vector w = (-1,1). The bias term is 0. The classifier makes a prediction during training to predict the class for the data point (-4,-6) and the dropout procedure sets the first component of this vector to 0. Which of the following statements is true?