Machine Learning and Neural Networks AIE Practice Questions (Page 7)

Questions 31–35

1. 31application · medium

   A developer is designing a neural network to classify images. They want to introduce non-linearity into the model so it can learn complex patterns. Which component is essential for this purpose?

   • AA linear transformation that only multiplies inputs by weights
   • BA loss function that measures the difference between predictions and labels
   • CAn activation function applied after the weighted sum in each neuron
   • DA learning rate that controls the step size during weight updates
   Reveal Solution Discussion Select an answer first
2. 32application · medium

   A retail bank wants to automatically approve or decline small personal loan applications based on historical applicant data (income, credit score, existing debt, loan amount) and the known outcome (approved or declined) for each past application. The model must output a clear yes/no decision for each new applicant. Which machine learning approach should the data science team use?

   • ASupervised regression using historical outcomes
   • BSupervised classification using labeled historical outcomes
   • CReinforcement learning with a reward for correct approvals
   • DUnsupervised clustering of applicants into risk groups
   Reveal Solution Discussion Select an answer first
4. 33foundation · easy

   In a neural network, what is the role of a hidden layer?

   • AIt stores the training labels for each example.
   • BIt receives the raw input data and passes it to the output layer.
   • CIt processes inputs and extracts features that are not directly visible in the input data.
   • DIt produces the final prediction or classification.
   Reveal Solution Discussion Select an answer first
5. 34application · medium

   A team is designing a neural network to classify images. They plan to use multiple hidden layers with a nonlinear activation function between them. Why is a nonlinear activation function necessary in this architecture?

   • ANonlinear activation functions speed up backpropagation by eliminating the need for gradient computation
   • BWithout nonlinearity, stacking multiple layers would collapse into a single linear transformation, limiting the model's ability to learn complex patterns
   • CNonlinear activation functions reduce the number of hidden layers needed to zero
   • DNonlinear activation functions ensure the network converges to the global minimum of the loss function
   Reveal Solution Discussion Select an answer first
6. 35expert · hard

   A team is training a deep neural network for image recognition. The network has many layers, but training is slow because the gradients in the early layers become very small, causing those layers to learn very slowly. Which technique is most effective for addressing this vanishing gradient problem?

   • ARemove all activation functions to allow gradients to flow freely
   • BUse activation functions like ReLU and apply batch normalization
   • CIncrease the learning rate significantly to compensate for small gradients
   • DReduce the number of layers to one hidden layer
   Reveal Solution Discussion Select an answer first