COELE1-2

1. "Mini-batch Gradient Descent" is often preferred because it:

A) Is only applicable to linear models.
B) Offers a balance between the efficiency of batch GD and the robustness of SGD.
C) Is guaranteed to converge faster than any other method.
D) Does not require a loss function.

2. "Transfer Learning" in deep learning involves:

A) Training a model from scratch on every new problem.
B) Using only unsupervised learning techniques.
C) Taking a model pre-trained on a large dataset (e.g., ImageNet) and fine-tuning it for a new, specific task with a smaller dataset.
D) Forgetting everything a model has learned.

3. An "Autoencoder" is a type of neural network primarily used for:

A) Supervised classification of images.
B) Predicting continuous values in a regression task.
C) Reinforcement learning.
D) Unsupervised learning tasks like dimensionality reduction and data denoising.b

4. The architecture of a typical autoencoder consists of:

A) A single output neuron with a linear activation.
B) Only a single layer of perceptrons.
C) An encoder that compresses the input and a decoder that reconstructs the input from the compression.
D) A convolutional layer followed by an RNN layer.

5. In the context of model evaluation for classification, "Accuracy" is defined as:

A) The proportion of total predictions that were correct.
B) The proportion of actual positives that were identified correctly.
C) The harmonic mean of precision and recall.
D) The proportion of positive identifications that were actually correct.

6. "Precision" is an important metric when:

A) The cost of false positives is high (e.g., in spam detection, where you don't want to flag legitimate emails as spam).
B) You are evaluating a regression model.
C) You need a single metric that combines precision and recall.
D) The cost of false negatives is high (e.g., in disease screening, where you don't want to miss a sick patient).

7. "Recall" is an important metric when:

A) The cost of false negatives is high (e.g., in disease screening, where you don't want to miss a sick patient).
B) The cost of false positives is high (e.g., in spam detection).
C) You need a single metric that combines precision and recall.
D) You are evaluating a clustering model.

8. The "F1 Score" is:

A) A metric used exclusively for regression.
B) The harmonic mean of precision and recall, providing a single score that balances both concerns.
C) The difference between precision and recall.
D) The arithmetic mean of precision and recall.

9. For a regression model, the "Mean Squared Error" (MSE) measures:

A) The total number of misclassified instances.
B) The variance of the input features.
C) The average of the squares of the errors between predicted and actual values.
D) The accuracy of a classification model.

10. The "ROC Curve" is a tool used to evaluate:

A) The architecture of a neural network.
B) The loss of a regression model over time.
C) The clustering quality of a K-means algorithm.
D) The performance of a binary classification model at various classification thresholds.

11. "Area Under the ROC Curve" (AUC) provides an aggregate measure of performance across all possible classification thresholds. A perfect model has an AUC of:

A) 0.0.
B) 0.5.
C) 1.0.
D) -1.0.

12. "K-fold Cross-Validation" is a technique used to:

A) Obtain a more robust estimate of model performance by training and evaluating the model K times on different splits of the data.
B) Visualize high-dimensional data.
C) Replace the need for a separate test set.
D) Increase the size of the training dataset.

13. In the K-Nearest Neighbors (K-NN) algorithm for classification, the class of a new data point is determined by:

A) A single, pre-defined rule.
B) A random selection from the training set.
C) The output of a linear function.u
D) The majority vote among its K closest neighbors in the feature space.

14. The parameter 'K' in the K-NN algorithm:

A) Is the learning rate for the algorithm.
B) Is always set to 1 for the best performance.
C) Is the number of features in the dataset.
D) Controls the model's flexibility. A small K can lead to overfitting, while a large K can lead to underfitting.

15. "Principal Component Analysis" (PCA) works by:

A) Classifying data using a decision boundary.
B) Predicting a target variable using linear combinations of features.
C) Finding new, uncorrelated dimensions (principal components) that capture the maximum variance in the data.u
D) Clustering data into K groups.

16. The first principal component in PCA is the direction in the feature space that:

A) Captures the greatest possible variance in the data.
B) Is perpendicular to all other components.
C) Captures the least possible variance in the data.
D) Is randomly oriented.

17. "K-Means Clustering" aims to partition data into K clusters such that:

A) The data is perfectly classified into known labels.
B) The between-cluster variance is minimized.
C) The data is projected onto a single dimension.
D) The within-cluster variance is minimized.

18. The "Elbow Method" is a heuristic used in K-Means to:

A) Help choose the optimal number of clusters K by looking for a "bend" in the plot of within-cluster variance.
B) Evaluate the accuracy of a classification model.
C) Initialize the cluster centroids.
D) Determine the learning rate for gradient descent.

19. "Naive Bayes" classifiers are called "naive" because they:

A) Make a strong (naive) assumption that all features are conditionally independent given the class label.
B) Always have the lowest possible accuracy.
C) Do not use probability in their predictions.
D) Are very simple and cannot handle complex data.

20. "Logistic Regression" is fundamentally a:

A) Clustering algorithm for grouping unlabeled data.
B) Regression algorithm for predicting continuous values.
C) Dimensionality reduction technique.
D) Classification algorithm that models the probability of a binary outcome using a logistic function.

21. The output of a logistic regression model is a value between 0 and 1, which represents the:

A) Probability that the input belongs to a particular class.
B) Exact value of the target variable.
C) Distance to the decision boundary.
D) Number of features in the input.

22. A "Random Forest" is an ensemble method that combines multiple:

A) Support Vector Machines.
B) Decision Trees to reduce overfitting and improve generalization.
C) K-NN models.
D) Linear Regression models.

23. The "bagging" technique in a Random Forest helps to:

A) Increase the speed of a single decision tree.
B) Perform feature extraction like PCA.
C) Reduce bias by making trees more complex.
D) Reduce variance by training individual trees on random subsets of the data and averaging their results

24. "Gradient Boosting" machines (e.g., XGBoost) are ensemble methods that:

A) Build all models independently and average them.
B) Do not require any parameter tuning.
C) Are exclusively used for unsupervised learning.
D) Build models sequentially, where each new model corrects the errors of the previous ones.

25. The term "feature engineering" refers to:

A) The evaluation of a model's final performance.
B) The process of using domain knowledge to create new input features that make machine learning algorithms work better.
C) The process of deleting all features from a dataset.
D) The automatic learning of features by a deep neural network.

26. "One-hot encoding" is a preprocessing technique used to:

A) Reduce the dimensionality of image data.
B) Convert categorical variables into a binary (0/1) format that can be provided to ML algorithms.
C) Normalize continuous numerical features.
D) Cluster similar data points together.

27. "Feature scaling" (e.g., normalization or standardization) is often crucial for algorithms that:

A) Are based on distance calculations or gradient descent, such as SVM and Neural Networks.
B) Are used for association rule learning.
C) Are used for clustering only.
D) Are based on tree-based models like Decision Trees and Random Forests.

28. The "curse of dimensionality" refers to the problem that:

A) There are never enough features to train a good model.
B) Dimensionality reduction always improves model performance.
C) As the number of features grows, the data becomes increasingly sparse, making it harder to find meaningful patterns.
D) All datasets should have as many features as possible.

29. "Regularization" is a technique used to:

A) Speed up the training time of a model.
B) Increase the variance of a model.
C) Make models more complex to fit the training data better.
D) Prevent overfitting by adding a penalty term to the loss function that discourages complex models.

30. L1 Regularization (Lasso) can often lead to:

A) Sparse models where the weights of less important features are driven to zero, effectively performing feature selection.
B) A decrease in model interpretability.
C) All features having non-zero weights.
D) Increased model complexity.

31. "Hyperparameters" are:

A) Configuration settings for the learning algorithm that are not learned from the data and must be set prior to training (e.g., learning rate, K in K-NN).
B) The input features of the model.
C) The parameters that the model learns during training (e.g., weights in a neural network).
D) The output predictions of the model.

32. The process of "Hyperparameter Tuning" involves:

A) Deploying the final model.
B) Training the model's internal weights.
C) Searching for the best combination of hyperparameters that results in the best model performance.
D) Cleaning the raw data.

33. "Grid Search" is a common method for hyperparameter tuning that involves:

A) Using a separate neural network to predict the best hyperparameters.
B) Ignoring hyperparameters altogether.
C) Exhaustively searching over a specified set of hyperparameter values.
D) Randomly sampling hyperparameter combinations from a distribution.

34. "Early Stopping" is a form of regularization that works by:

A) Starting the training process later than scheduled.
B) Stopping the training after a fixed, very short number of epochs.
C) Halting the training process when performance on a validation set starts to degrade, indicating the onset of overfitting.
D) Using a very small learning rate.

35. A "Vanilla" neural network, also known as a Multilayer Perceptron (MLP), is typically composed of:

A) A single layer of neurons.
B) Fully connected layers, where each neuron in one layer is connected to every neuron in the next layer.
C) Recurrent layers for processing sequences.
D) Convolutional layers for processing images.

36. The "softmax" activation function is commonly used in the output layer of a neural network for:

A) Unsupervised learning problems.
B) Binary classification problems.
C) Multi-class classification problems, as it outputs a probability distribution over the possible classes.
D) Regression problems.

37. The "Adam" optimizer is an adaptive learning rate algorithm that is often preferred because it:

A) Does not require any hyperparameters.
B) Is guaranteed to find the global minimum for any function.
C) Is only used for unsupervised learning.
D) Combines the advantages of two other extensions of stochastic gradient descent, AdaGrad and RMSProp.

38. "Batch Normalization" is a technique used to:

A) Replace the need for an activation function.
B) Increase the batch size during training.
C) Improve the stability and speed of neural network training by normalizing the inputs to each layer.
D) Normalize the entire dataset before feeding it into the network.

39. The "confusion matrix" is a table that is used to describe the performance of a:

A) Clustering algorithm's group assignments.
B) Dimensionality reduction technique's effectiveness.
C) Regression model's accuracy.
D) Classification model on a set of test data for which the true values are known.

40. In a confusion matrix, the "true positives" are the cases where:

A) The model correctly predicted the negative class.
B) The model incorrectly predicted the positive class.
C) The model incorrectly predicted the negative class.
D) The model correctly predicted the positive class.

41. The problem of "imbalanced classes" occurs when:

A) The model is too complex for the data.
B) One class in the training data has significantly more examples than another, which can bias the model.
C) The features are not scaled properly.
D) The learning rate is set too high.

42. A technique to address imbalanced classes is "SMOTE," which:

A) Combines all classes into one.
B) Generates synthetic examples for the minority class to balance the dataset.
C) Deletes examples from the majority class at random.
D) Ignores the minority class completely.

43. "Reinforcement Learning" differs from supervised and unsupervised learning in that:

A) It is a simpler and less powerful approach.
B) It requires a fully labeled dataset for training.
C) It is only used for clustering unlabeled data.
D) It learns by interacting with an environment and receiving rewards or penalties for actions, without a labeled dataset.

44. "Q-Learning" is a popular algorithm in reinforcement learning that learns:

A) The principal components of a state space.
B) A decision tree for classification.
C) A clustering of possible actions.
D) A policy that tells an agent what action to take under what circumstances by learning a value function.

45. "Natural Language Processing" (NLP) often uses supervised learning for tasks like:

A) Grouping similar news articles without labels.
B) Generating new, original text without any input.
C) Sentiment analysis, where text is classified as positive, negative, or neutral.
D) Reducing the dimensionality of word vectors.

46. "Word Embeddings" (like Word2Vec) are techniques that:

A) Are used only for image classification.
B) Represent words as dense vectors in a continuous space, capturing semantic meaning.
C) Represent words as simple one-hot encoded vectors.
D) Are a type of clustering algorithm.

47. A "Generative Adversarial Network" (GAN) consists of two networks:

A) A Generator and a Discriminator, which are trained in opposition to each other.
B) An Encoder and a Decoder for compression.
C) Two identical Convolutional Neural Networks.
D) A single, large Regression network.

48. The "Generator" in a GAN is responsible for:

A) Reducing the dimensionality of the input.
B) Classifying input images into categories.
C) Discriminating between real and fake data.
D) Creating new, synthetic data that is indistinguishable from real data.

49. The "Discriminator" in a GAN is essentially a:

A) Regression model predicting a continuous value.
B) Dimensionality reduction technique.
C) Clustering algorithm grouping similar images.
D) Binary classifier that tries to correctly label data as real (from the dataset) or fake (from the generator).

Created with That Quiz — a math test site for students of all grade levels.