🏠 Home
← Back Start Quiz
MCQ Practice

Technological Knowledge MCQs

Which loss function is typically used for binary classification?

  • A. Softmax loss
  • B. Mean squared error
  • C. Cross-entropy loss
  • D. Huber loss
Explanation:
Cross-entropy loss (binary or logistic) is commonly used for binary classification tasks.

Which concept refers to a model's ability to perform well on unseen data?

  • A. Underfitting
  • B. Overfitting
  • C. Generalization
  • D. Normalization
Explanation:
Generalization is a model's ability to apply learned patterns to new data.

What distinguishes 'batch' training from 'online' training?

  • A. Batch uses all data at once per update; online uses one sample at a time
  • B. Batch normalizes data; online does not
  • C. Online uses GPU; batch uses CPU
  • D. Batch increases learning rate automatically
Explanation:
Batch training updates using the entire dataset per iteration; online (stochastic) updates per sample.

What is the purpose of a confusion matrix?

  • A. To store model weights
  • B. To summarize classification performance
  • C. To distribute data across GPUs
  • D. To reduce dimensionality
Explanation:
A confusion matrix shows true vs predicted class counts, summarizing classification performance.

In deep learning, what is 'fine-tuning'?

  • A. Re-initializing weights
  • B. Training only the final layer of a pre-trained model
  • C. Decreasing model depth
  • D. Hyperparameter optimization
Explanation:
Fine-tuning means training or adjusting a pre-trained model (often its final layers) on new data.

Which of these is a type of unsupervised learning algorithm?

  • A. Decision tree
  • B. K-means clustering
  • C. Logistic regression
  • D. Random forest
Explanation:
K-means is a clustering algorithm used in unsupervised learning.

What is a common activation function used in the output layer for multi-class classification?

  • A. Sigmoid
  • B. Softmax
  • C. ReLU
  • D. Linear
Explanation:
Softmax activation is used for multi-class outputs to produce probabilities summing to 1.

Which evaluation metric is most appropriate for an imbalanced classification dataset?

  • A. Accuracy
  • B. Precision/Recall or F1-score
  • C. Mean squared error
  • D. AUC (Area under ROC)
Explanation:
Precision, recall, or F1-score better reflect performance on imbalanced classes than accuracy.

What is the vanishing gradient problem?

  • A. Loss function failing to converge
  • B. Gradients are becoming very small in deep networks
  • C. Model accuracy exceeding training accuracy
  • D. Optimization getting stuck in local minima
Explanation:
The vanishing gradient problem occurs when gradients shrink, hampering learning in deep networks.

What technique helps mitigate overfitting by adding a penalty on large weights?

  • A. Regularization (like L2)
  • B. Data augmentation
  • C. Increasing learning rate
  • D. Batch normalization
Explanation:
Regularization adds a cost for large weights to prevent overfitting.