Anomaly Detection and Optimization GMLE Practice Questions (Page 1)

Questions 1–5

1. 1application · medium

   A deep learning engineer is training a neural network with a large dataset. The training loss is decreasing, but the validation loss is not decreasing. The engineer suspects the model is not generalizing well. Which optimization strategy is most appropriate?

   • AIncrease the learning rate to speed up convergence.
   • BIncrease the batch size to reduce the noise in gradient estimates.
   • CSwitch from Adam to SGD with momentum.
   • DApply dropout regularization to the hidden layers.
   Reveal Solution Discussion Select an answer first
2. 2foundation · easy

   In anomaly detection, what is the primary purpose of adjusting the decision threshold?

   • ATo change the learning rate of the optimization algorithm
   • BTo select which features are used in the model
   • CTo balance the trade-off between false positives and false negatives based on the business context
   • DTo determine the number of layers in a neural network
   Reveal Solution Discussion Select an answer first
4. 3expert · hard

   A sensor network monitors temperature readings. The data is known to have occasional sensor failures that produce extreme values, but the underlying process also has natural fluctuations. The team wants to distinguish between sensor failures and genuine process anomalies. They have a small set of labeled failure events. Which approach is most appropriate?

   • ATrain a supervised classifier on the labeled failures and normal data.
   • BUse a z-score with a high threshold to flag only extreme values.
   • CUse a Gaussian mixture model with two components.
   • DUse an isolation forest on all data and flag the most anomalous points.
   Reveal Solution Discussion Select an answer first
5. 4foundation · easy

   In the context of anomaly detection, what is the primary purpose of an autoencoder?

   • ATo reduce the dimensionality of the data so that anomalies become more visible in a lower-dimensional space
   • BTo classify data points into normal and anomalous classes using a softmax output layer
   • CTo generate synthetic anomalies that can be used to train a supervised classifier
   • DTo learn a compressed representation of the input data and reconstruct it, with high reconstruction error indicating an anomaly
   Reveal Solution Discussion Select an answer first
6. 5expert · hard

   A data scientist is building an anomaly detection model for a manufacturing process. The data is highly imbalanced, with only 0.1% of samples being anomalous. The scientist trains a model and evaluates it using ROC-AUC, achieving a score of 0.95. However, when the model is deployed, it flags a large number of false positives. What is the most likely reason for this discrepancy?

   • AThe model is not calibrated, and the predicted probabilities do not reflect the true likelihood of anomaly.
   • BROC-AUC is not a suitable metric for imbalanced datasets because it is insensitive to the minority class.
   • CThe model is underfitting, and the ROC-AUC is artificially high due to the class imbalance.
   • DThe model is overfitting to the training data, causing it to perform poorly on new data.
   Reveal Solution Discussion Select an answer first