AI0-001 AI Concepts and Techniques Practice Test 4 — 15 Questions

Question 1

A machine learning team is splitting a dataset for a binary classification problem. They want to ensure robust evaluation and avoid data leakage. Which TWO practices should they follow? (Choose 2)

Accepted Answer

Split into training, validation, and test sets. Train/validation/test split is standard; cross-validation gives more robust estimates. Shuffling before split prevents ordering bias.

Answer

Normalise the entire dataset before splitting

Answer

Include validation data in the training set for more data

Answer

Use the same split for all experiments

Question 2

A data scientist is building a model to predict credit default using historical loan data. The dataset contains 100,000 records with 50 features, including income, debt-to-income ratio, and loan amount. The target variable is binary (default vs. no default). The goal is to maximize interpretability while maintaining high accuracy. Which algorithm is MOST appropriate?

Accepted Answer

Logistic regression. Logistic regression is interpretable (coefficients show feature impact) and performs well on binary classification with a large dataset. Decision trees are interpretable but may overfit; random forests and gradient boosting are less interpretable.

Answer

Random forest

Answer

Gradient boosting machine

Answer

Decision tree

Question 3

A team is training a deep learning model for image classification. The training loss decreases steadily but the validation loss plateaus after 20 epochs and then starts to increase. Which action is MOST likely to improve generalization?

Accepted Answer

Implement early stopping. Early stopping halts training when validation loss stops improving, preventing overfitting. Increasing learning rate would worsen divergence; adding more layers increases capacity and overfitting; reducing batch size may help optimization but not directly address overfitting.

Answer

Add more convolutional layers

Answer

Increase the learning rate

Answer

Reduce the batch size

Question 4

Which machine learning paradigm is best suited for training a model to play a game by learning from its own actions and rewards, without labeled data?

Accepted Answer

Reinforcement learning. Reinforcement learning learns via trial-and-error using rewards and penalties, ideal for game-playing agents. Supervised learning requires labeled data; unsupervised learning finds patterns without rewards; semi-supervised uses a mix.

Answer

Unsupervised learning

Answer

Semi-supervised learning

Answer

Supervised learning

Question 5

A natural language processing team wants to build a sentiment analysis model for customer reviews. They have 10,000 labeled reviews and 1 million unlabeled reviews. Which approach would MOST effectively leverage the unlabeled data?

Accepted Answer

Implement a semi-supervised learning algorithm that propagates labels from the labeled to the unlabeled data. Semi-supervised learning uses the small labeled set to guide learning from the large unlabeled set. Self-supervised learning would require a pretext task; fine-tuning a pre-trained model is also valid but semi-supervised directly addresses the labeled-unlabeled mix.

Answer

Use self-supervised learning to pretrain on the unlabeled data, then fine-tune on the labeled data

Answer

Train a supervised classifier on only the 10,000 labeled reviews

Answer

Use reinforcement learning with the unlabeled data as rewards

Question 6

A company is deploying a text generation model for customer service emails. They want to ensure the model's responses are factual and based on internal knowledge bases. Which technique is most effective?

Accepted Answer

Use Retrieval-Augmented Generation (RAG). RAG retrieves relevant documents from the knowledge base at inference time, grounding the model's responses in facts. Prompt engineering alone can't ensure factual accuracy; fine-tuning may still hallucinate; temperature reduction only reduces randomness.

Answer

Fine-tune the model on historical customer service emails

Answer

Write a detailed system prompt

Answer

Set the temperature to 0

Question 7

A machine learning engineer is training a transformer model for machine translation. The model's perplexity on the validation set is 8.5, and the BLEU score is 32. After increasing the number of encoder layers from 6 to 12, perplexity drops to 7.2 but BLEU decreases to 28. What is the MOST likely cause?

Accepted Answer

The model is overfitting the training data. Perplexity measures language model confidence, but BLEU measures translation quality. The deeper model may overfit to the training data, reducing perplexity but hurting generalization to validation translations. Overfitting causes high confidence (low perplexity) but poor translation diversity or exact matches.

Answer

The batch size is too small

Answer

The model is underfitting the training data

Answer

The learning rate is too high

Question 8

Which neural network architecture is specifically designed to handle sequential data and mitigate the vanishing gradient problem?

Accepted Answer

Long Short-Term Memory (LSTM) network. LSTM (Long Short-Term Memory) is a type of RNN designed with gating mechanisms to prevent vanishing gradients in long sequences. CNNs are for spatial data; vanilla RNNs suffer from vanishing gradients; transformers use attention but are not specifically designed to mitigate vanishing gradients (they use residual connections).

Answer

Convolutional Neural Network (CNN)

Answer

Transformer

Answer

Vanilla Recurrent Neural Network (RNN)

Question 9

A data scientist is evaluating a binary classification model. The model achieves 95% accuracy on the test set, but the precision is 0.60 and recall is 0.55. The dataset has 90% negative class samples. Which metric should the team focus on to improve the model?

Accepted Answer

F1 score. With high class imbalance (90% negatives), accuracy is misleading. F1 score balances precision and recall, giving a better picture of performance on the minority class. AUC-ROC is also good but F1 directly optimizes for positive class.

Answer

Perplexity

Answer

BLEU score

Answer

Accuracy

Question 10

A company wants to generate realistic images of new product designs. They have a large dataset of existing product images. Which generative AI approach is MOST suitable for creating novel, high-quality images?

Accepted Answer

Generative adversarial network (GAN). GANs (Generative Adversarial Networks) consist of a generator and discriminator that compete, producing highly realistic images. Diffusion models are also good but GANs are historically the go-to for image generation. VAEs produce blurrier images; LLMs are for text.

Answer

Large language model (LLM)

Answer

Variational autoencoder (VAE)

Answer

Diffusion model

Question 11

A team is fine-tuning a BERT model for a document classification task. They notice the model achieves high F1 scores on the training set but low F1 on the validation set. Which regularization technique would be MOST effective?

Accepted Answer

Dropout. Dropout randomly drops neurons during training, preventing co-adaptation and overfitting. L1 and L2 add penalties to weights but are less common for transformers; L1 induces sparsity, L2 reduces weight magnitude. However, dropout is the standard regularization in BERT-like models.

Answer

L1 regularization

Answer

L2 regularization

Answer

Reduce batch size

Question 12

In unsupervised learning, which task involves grouping similar data points together based on feature similarities?

Accepted Answer

Clustering. Clustering partitions data into groups where intra-cluster similarity is high. Classification is supervised; anomaly detection finds outliers; regression predicts continuous values.

Answer

Anomaly detection

Answer

Classification

Answer

Regression

Question 13

A developer is using a large language model via an API. They want the model to solve a math problem step by step. Which prompt engineering technique should they use?

Accepted Answer

Chain-of-thought prompting. Chain-of-thought prompting encourages the model to show intermediate reasoning steps, improving accuracy on multi-step problems. Zero-shot gives no examples; few-shot provides examples but not necessarily step-by-step; temperature controls randomness.

Answer

Set temperature to 0.9

Answer

Few-shot prompting

Answer

Zero-shot prompting

Question 14

A company wants to deploy an LLM-based chatbot that can handle sensitive customer information. Which THREE measures should be implemented to mitigate prompt injection attacks? (Choose 3)

Accepted Answer

Use a system prompt that instructs the model to ignore any instructions in the user input. Input sanitization removes special characters or patterns; output filtering checks responses for sensitive data; system prompts with separation instructions can reduce injection risk. Restricting temperature only affects randomness; using a smaller model does not prevent injection.

Answer

Use a smaller model with fewer parameters

Answer

Set temperature to a low value

Question 15

A machine learning team is evaluating a logistic regression model for a binary classification task. The dataset has 1,000 samples and 20 features. Which TWO metrics are most appropriate for evaluating model performance? (Choose 2)

Accepted Answer

F1 score. The F1 score is appropriate because it balances precision and recall, making it robust for binary classification when class distribution may be imbalanced. AUC-ROC measures the model's ability to distinguish between positive and negative classes across all classification thresholds, providing a threshold-independent evaluation of discriminative performance.

Answer

BLEU score

Answer

Perplexity

Answer

Accuracy