Machine Learning and Neural Networks AIE Practice Questions (Page 9)

Questions 41–45

1. 41expert · hard

   A financial firm wants to build a system that recommends personalized investment portfolios. The system should learn from each client's past investment decisions and feedback, and adapt over time as client preferences change. The firm also wants to ensure the system can explain its recommendations to comply with regulations. Which approach is most appropriate?

   • AReinforcement learning with a reward function based on client satisfaction and portfolio performance, combined with interpretable models or post-hoc explanations
   • BUnsupervised clustering of clients to create static portfolio recommendations
   • CA fixed rule-based system that follows predefined investment guidelines
   • DSupervised learning on historical portfolio returns to predict future performance
   Reveal Solution Discussion Select an answer first
2. 42application · medium

   A marketing team has a large dataset of customer transactions with hundreds of features but no predefined categories or labels. They want to identify natural segments of customers for targeted campaigns and also reduce the number of features to simplify visualization. Which approach should they use?

   • ABuild a deep neural network with multiple hidden layers to output customer segment labels directly
   • BApply k-means clustering to segment customers and then use principal component analysis (PCA) to reduce dimensionality for visualization
   • CTrain a supervised classifier on the customer data to predict segment membership
   • DUse a reinforcement learning agent to explore customer groupings and maximize a reward function
   Reveal Solution Discussion Select an answer first
4. 43application · medium

   During backpropagation, the network computes gradients of the loss with respect to the weights. What is the primary purpose of these gradients?

   • ATo determine the number of hidden layers needed
   • BTo update the weights in the direction that reduces the loss
   • CTo calculate the accuracy of the model on the training set
   • DTo select the activation function for each neuron
   Reveal Solution Discussion Select an answer first
5. 44application · medium

   A healthcare startup wants to predict the likelihood that a patient will be readmitted within 30 days. They have historical patient records with known readmission outcomes. Which machine learning type should they use, and why?

   • ASupervised learning, because the historical records include the known outcome (readmitted or not) to train the model
   • BUnsupervised learning, because the model should discover patterns in patient data without predefined outcomes
   • CReinforcement learning, because the model should learn from the consequences of treatment decisions
   • DTraditional programming, because readmission rules can be manually coded from medical guidelines
   Reveal Solution Discussion Select an answer first
6. 45foundation · easy

   Which statement best describes unsupervised learning?

   • AIt finds patterns or structure in data that has no labeled outputs.
   • BIt learns by interacting with an environment and receiving rewards.
   • CIt uses a teacher to correct the model's mistakes during training.
   • DIt learns to map inputs to known outputs using labeled examples.
   Reveal Solution Discussion Select an answer first