AI0-001 Machine Learning and Deep Learning Practice Test 5 — 15 Questions

Question 1

An AI engineer is training a deep neural network for image recognition. The training loss decreases steadily for the first few epochs but then plateaus and starts to oscillate. Which adjustment is most likely to improve convergence?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Reduce the learning rate. The plateau and oscillation of the training loss indicate that the optimizer is overshooting the minimum due to a learning rate that is too high. Reducing the learning rate allows the optimizer to take smaller, more precise steps, dampening oscillations and enabling convergence to a lower loss. This is a standard technique in gradient descent optimization, often implemented via learning rate schedules or adaptive methods like Adam.

Answer

Add more layers

Answer

Increase the learning rate

Answer

Increase the batch size

Question 2

A healthcare organization wants to use patient data to predict disease risk. They are concerned about bias in the model. Which step is most critical during the data preparation phase to mitigate bias?

Accepted Answer

Ensuring the training data is representative of the target population. Ensuring the training data is representative of the target population is the most critical step during data preparation to mitigate bias because bias often originates from skewed or incomplete data that does not reflect the real-world distribution of patient demographics, conditions, and outcomes. Without a representative dataset, any subsequent preprocessing or modeling will propagate and potentially amplify existing disparities, leading to unfair or inaccurate predictions for underrepresented groups.

Answer

Applying SMOTE to oversample minority classes

Answer

Using a more complex algorithm

Answer

Removing all demographic features

Question 3

A deep learning model for natural language processing uses a recurrent neural network (RNN) to process long sequences. The gradients vanish after many time steps. Which architectural change is most effective to mitigate this problem?

Accepted Answer

Replace the RNN cells with Long Short-Term Memory (LSTM) units. LSTMs are specifically designed with a gating mechanism (input, forget, and output gates) and a cell state that allows gradients to flow unchanged over many time steps, directly addressing the vanishing gradient problem in standard RNNs. This architectural change preserves long-range dependencies in sequences, which is critical for tasks like language modeling or machine translation.

Answer

Add dropout regularization

Answer

Use a larger learning rate

Answer

Increase the number of hidden layers

Question 4

An organization has a dataset with categorical features having high cardinality (e.g., ZIP codes). They plan to use a tree-based model. Which encoding method is most appropriate?

Accepted Answer

Target encoding (mean encoding). Target encoding (mean encoding) replaces each category with the mean of the target variable for that category, which works well with tree-based models on high-cardinality features because it captures the predictive signal without exploding the feature space. This method avoids the dimensionality explosion of one-hot encoding and the arbitrary ordering of label encoding, making it the most appropriate choice for high-cardinality categorical features in tree-based models.

Answer

Label encoding

Answer

One-hot encoding

Answer

Frequency encoding

Question 5

A company deploys a machine learning model that makes predictions on streaming data. Over time, the data distribution shifts, causing model performance to degrade. Which monitoring strategy is most appropriate to detect this drift?

Accepted Answer

Track the model's accuracy on a fixed validation set over time. Tracking the model's accuracy on a fixed validation set over time directly measures performance degradation caused by data drift. As the streaming data distribution shifts, the model's predictions on the static validation set will become less accurate, providing a clear signal that drift has occurred. This is a standard monitoring approach in production ML systems for detecting concept drift.

Answer

Compare the distribution of predictions to the training set

Answer

Monitor the model's training loss

Answer

Retrain the model daily on new data

Question 6

A data scientist is tuning hyperparameters for a support vector machine (SVM) with an RBF kernel. Which two hyperparameters most significantly affect model performance? (Select TWO.)

Accepted Answer

gamma (kernel coefficient). Gamma defines the influence of a single training example, with low values meaning a far reach and high values meaning a close reach. C controls the trade-off between achieving a low error on the training data and minimizing the margin, directly impacting overfitting. Together, they are the two most critical hyperparameters for an SVM with an RBF kernel.

Answer

learning rate

Answer

epsilon (for epsilon-SVR)

Answer

degree (for polynomial kernel)

Question 7

A team is designing a deep learning pipeline for a computer vision task. They want to reduce overfitting. Which two techniques are specifically effective for this purpose? (Select TWO.)

Accepted Answer

Dropout. Dropout is a regularization technique that randomly drops a fraction of neurons during training, which prevents the network from relying too heavily on any single neuron and forces it to learn more robust features. This reduces overfitting by introducing noise that improves generalization.

Answer

Using a smaller batch size

Answer

Adding more layers

Answer

Increasing the learning rate

Question 8

Refer to the exhibit. The training log shows losses and accuracies over 5 epochs. What is the most likely problem?

Accepted Answer

Overfitting. The training log shows high training accuracy (e.g., 99%) but low validation accuracy (e.g., 60%) across epochs, with the validation loss increasing after an initial drop. This divergence indicates the model has memorized the training data rather than learning generalizable patterns, which is the hallmark of overfitting.

Answer

Data leakage

Answer

Underfitting

Answer

Vanishing gradient

Question 9

A data scientist needs to predict whether a customer will churn based on historical data containing features like account age, monthly charges, and support tickets. The target variable is binary (churn or not). Which type of machine learning algorithm should be used?

Accepted Answer

Logistic regression. Logistic regression is the correct choice because it is specifically designed for binary classification tasks, such as predicting whether a customer will churn (yes/no). It models the probability of the binary outcome using a logistic (sigmoid) function, making it suitable for this supervised learning problem with a categorical target variable.

Answer

Linear regression

Answer

K-means clustering

Answer

Principal component analysis

Question 10

A team trained a deep neural network on a limited dataset. The training loss decreases consistently, but the validation loss starts increasing after 20 epochs. What is the most likely issue and the best corrective action?

Accepted Answer

Overfitting; apply regularization like dropout. The training loss decreasing while validation loss increasing after 20 epochs is the classic signature of overfitting: the model has memorized the training data but fails to generalize to unseen data. Applying regularization like dropout forces the network to learn more robust features by randomly dropping neurons during training, reducing overfitting. This is the most direct and effective corrective action for this specific symptom.

Answer

Vanishing gradient; use ReLU activation

Answer

Underfitting; increase model complexity

Answer

Data leakage; reshuffle split

Question 11

A company is building a computer vision system to detect defects in manufactured parts. They have 10,000 labeled images per class (defective and non-defective). They want to achieve high accuracy with limited computational resources. Which deep learning architecture and approach is most appropriate?

Accepted Answer

Use a pre-trained VGG16 and fine-tune the last few layers. Using a pre-trained VGG16 and fine-tuning the last few layers leverages transfer learning, which is ideal when you have a moderate-sized labeled dataset (10,000 images per class) and limited computational resources. The pre-trained model already captures general visual features from ImageNet, so only the task-specific layers need to be trained, reducing training time and resource requirements while still achieving high accuracy.

Answer

Train a custom CNN from scratch with many layers

Answer

Use a decision tree ensemble

Answer

Use an RNN to process image sequences

Question 12

A machine learning engineer has a dataset of 100,000 records. She splits it into 70% training, 15% validation, and 15% test sets. After training, the model achieves 95% accuracy on training and 85% on validation. What does the accuracy difference most likely indicate?

Accepted Answer

The model is overfitting. The 10% gap between training accuracy (95%) and validation accuracy (85%) is a classic sign of overfitting. The model has memorized patterns specific to the training set rather than learning generalizable features, causing it to perform worse on unseen validation data. In machine learning, a significant drop in performance from training to validation indicates poor generalization, which is the hallmark of overfitting.

Answer

The validation set is too small

Answer

The model generalizes well

Answer

The test set should be larger

Question 13

A deep learning model for sentiment analysis uses a softmax output layer. The hidden layers currently use tanh activation. Which activation function should replace tanh to mitigate vanishing gradients in deeper networks?

Accepted Answer

ReLU. ReLU (Rectified Linear Unit) is correct because it outputs zero for negative inputs and a positive linear slope for positive inputs, which avoids the saturation problem of tanh. In deeper networks, tanh gradients can vanish as activations approach ±1, slowing or halting learning. ReLU's non-saturating nature keeps gradients flowing for positive inputs, mitigating the vanishing gradient problem.

Answer

Sigmoid

Answer

Softmax

Answer

Linear

Question 14

A fraud detection model is trained on a dataset where only 0.1% of transactions are fraudulent. The model achieves 99.9% accuracy but fails to catch most frauds. Which metric should the team prioritize, and which technique could help?

Accepted Answer

Precision-Recall AUC; use oversampling like SMOTE. The dataset is highly imbalanced (0.1% fraud), so 99.9% accuracy is misleading because a model that predicts 'not fraud' for every transaction achieves it. Precision-Recall AUC focuses on the positive class (fraud) and is robust to class imbalance, unlike accuracy or ROC-AUC. Oversampling like SMOTE generates synthetic fraud samples to balance the dataset, helping the model learn the minority class patterns.

Answer

Mean Squared Error; use L2 regularization

Answer

F1 score; use principal component analysis

Answer

Accuracy; collect more data

Question 15

A dataset contains features on vastly different scales (e.g., age 0-100 vs. income 0-1,000,000). Which preprocessing step is essential before training a neural network?

Accepted Answer

Feature scaling (standardization or normalization). Neural networks rely on gradient-based optimization, where features with larger scales can dominate the weight updates, causing unstable convergence or slow training. Feature scaling (standardization or normalization) ensures all features contribute equally to the loss function, preventing the model from being biased toward high-magnitude features like income versus age.

Answer

Data augmentation

Answer

Dimensionality reduction

Answer

One-hot encoding