Anomaly Detection and Optimization GMLE Practice Questions (Page 9)

Questions 41–45

1. 41application · medium

   A data scientist is training a deep neural network for image classification. The training loss is decreasing, but the validation loss starts to increase after 20 epochs. The scientist wants to improve generalization without changing the model architecture. Which action should they take?

   • AApply early stopping, saving the model weights from the epoch with the lowest validation loss.
   • BIncrease the batch size to reduce the noise in gradient estimates.
   • CSwitch from Adam to SGD with momentum to improve convergence.
   • DIncrease the learning rate to escape the local minimum.
   Reveal Solution Discussion Select an answer first
2. 42application · medium

   A manufacturing plant monitors the temperature of a furnace. The temperature is normally stable at 200°C with a standard deviation of 2°C. A quality engineer observes a reading of 206°C, which is 3 standard deviations above the mean. However, the engineer knows the furnace was intentionally set to a higher temperature for a new batch of material. How should the engineer classify this reading?

   • AAs a collective anomaly, because the sequence of readings over time will show a pattern shift.
   • BAs a point anomaly, because the value is statistically far from the historical mean.
   • CAs a contextual anomaly, because the value is unusual for the normal operating mode but expected for the new batch setting.
   • DAs a normal reading, because the z-score is only 3 and the process is known to be stable.
   Reveal Solution Discussion Select an answer first
4. 43foundation · easy

   Which hyperparameter optimization technique evaluates all possible combinations of a predefined set of hyperparameter values?

   • ARandom search
   • BEarly stopping
   • CGrid search
   • DBayesian optimization
   Reveal Solution Discussion Select an answer first
5. 44foundation · easy

   Which regularization technique involves monitoring the model's performance on a validation set and stopping training when performance begins to degrade?

   • AL1 regularization
   • BDropout
   • CEarly stopping
   • DBatch normalization
   Reveal Solution Discussion Select an answer first
6. 45expert · hard

   A deep learning model for speech recognition is training very slowly. The loss decreases but at a decreasing rate, and the gradient norms are consistently small. The team suspects the architecture has many layers with saturating activation functions. Which optimization strategy is most likely to improve convergence speed?

   • AIncrease the learning rate significantly.
   • BReplace saturating activations with ReLU and use batch normalization.
   • CUse only SGD without momentum.
   • DAdd more layers to increase model capacity.
   Reveal Solution Discussion Select an answer first