Multinomial Logistic Regression and Maximum Entropy DATA-SCIENCE-OPTIMIZE Practice Questions (Page 3)

Questions 11–15

1. 11application · medium

   A data scientist is training a multinomial logistic regression model with 5 classes and 100 features. They are using gradient descent with a fixed learning rate. After the first few iterations, the loss is decreasing, but the gradient norm is very large, causing the parameter updates to overshoot and the loss to oscillate. Which adjustment is most appropriate to stabilize training?

   • ASwitch to a different loss function, such as mean squared error, to reduce gradient magnitude
   • BAdd more features to the model to smooth the loss landscape
   • CDecrease the learning rate to reduce the size of each parameter update
   • DIncrease the learning rate to escape the oscillating region faster
   Reveal Solution Discussion Select an answer first
2. 12application · medium

   A data scientist is training a multinomial logistic regression model on a dataset with 3 classes and 50 features. They are using a batch gradient descent optimizer. After 100 iterations, the loss is still decreasing slowly. The data scientist wants to speed up convergence. Which change is most likely to help?

   • AUse a fixed learning rate that is larger than the current one
   • BIncrease the number of features to provide more information to the model
   • CUse a momentum-based optimizer, such as SGD with momentum, to accelerate convergence
   • DReduce the number of classes from 3 to 2 to simplify the problem
   Reveal Solution Discussion Select an answer first
4. 13foundation · easy

   How does the maximum entropy principle relate to multinomial logistic regression?

   • AIt provides a theoretical justification for the form of the multinomial logistic regression model
   • BIt replaces the need for training data in logistic regression
   • CIt is an alternative model that does not use feature constraints
   • DIt is used only for binary classification, not multi-class
   Reveal Solution Discussion Select an answer first
5. 14foundation · easy

   In multinomial logistic regression, what is the primary purpose of the softmax function?

   • ATo convert probabilities into binary class labels for each class independently
   • BTo transform raw class scores into a probability distribution that sums to 1
   • CTo scale features to have zero mean and unit variance before classification
   • DTo rank classes by their raw scores without changing their relative order
   Reveal Solution Discussion Select an answer first
6. 15foundation · easy

   In a maximum entropy model, how are binary features typically represented?

   • AAs continuous values that can be any real number
   • BAs categorical labels that are encoded as strings
   • CAs indicator functions that return 1 if a condition holds and 0 otherwise
   • DAs probabilities that sum to 1 across all features
   Reveal Solution Discussion Select an answer first