Machine Learning and Neural Networks AIE Practice Questions (Page 8)

Questions 36–40

1. 36application · easy

   A bank wants to approve or reject loan applications. They have historical data with features like income, credit score, and loan amount, along with the outcome (approved/rejected). They want to build a model that predicts the approval decision for new applicants. Which machine learning task is this?

   • ASupervised classification, because the target is a discrete category (approved or rejected)
   • BUnsupervised clustering, because the model groups similar loan applications
   • CReinforcement learning, because the model learns from loan approval rewards
   • DSupervised regression, because the model predicts a continuous approval score
   Reveal Solution Discussion Select an answer first
2. 37expert · medium

   A researcher is designing a neural network for a regression task. They notice that the model's predictions are always positive, but the true target values can be negative. Which activation function in the output layer is most appropriate?

   • AReLU activation, because it is fast and simple
   • BSigmoid activation, because it squashes outputs between 0 and 1
   • CLinear activation, because it allows the output to be any real number
   • DTanh activation, because it outputs between -1 and 1
   Reveal Solution Discussion Select an answer first
4. 38foundation · easy

   What distinguishes deep learning from traditional machine learning?

   • ADeep learning uses only unsupervised learning, while traditional machine learning uses supervised learning.
   • BDeep learning requires labeled data, while traditional machine learning does not.
   • CDeep learning uses neural networks with multiple hidden layers to learn hierarchical representations.
   • DDeep learning does not require feature engineering, while traditional machine learning always does.
   Reveal Solution Discussion Select an answer first
5. 39foundation · easy

   What is the purpose of a loss function in neural network training?

   • ATo add randomness to the training process to avoid overfitting.
   • BTo measure how far the network's predictions are from the true targets.
   • CTo determine the architecture of the network, such as the number of layers.
   • DTo store the training data in memory for faster access.
   Reveal Solution Discussion Select an answer first
6. 40expert · hard

   A company wants to deploy a deep learning model for real-time language translation on mobile devices. The model must run with limited memory and battery, but also maintain high accuracy. Which approach best balances these constraints?

   • AUse a very deep network with many layers to maximize accuracy, regardless of size
   • BRun the model in the cloud and stream results to the device
   • CTrain a large deep model, then use model compression techniques such as pruning and quantization to reduce its size
   • DUse a simple linear model to minimize resource usage
   Reveal Solution Discussion Select an answer first