AI0-001 Practice Test 20 — 15 Questions

Question 1

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

Question 2

A machine learning engineer is building a spam filter. The dataset contains 10,000 emails, of which 1,000 are spam. The engineer decides to use a Random Forest classifier. Which preprocessing step is most critical to ensure the model generalizes well to new, unseen emails?

Accepted Answer

Split the data into training and testing sets before any other preprocessing. Splitting the data into training and testing sets before any other preprocessing prevents data leakage. If preprocessing like normalization or PCA is applied to the entire dataset first, the test set information influences the training process, leading to overly optimistic performance estimates and poor generalization to new, unseen emails.

Answer

Apply Principal Component Analysis (PCA) to reduce dimensionality

Answer

Normalize the numerical features to have zero mean and unit variance

Answer

Encode all features using one-hot encoding

Question 3

An AI developer needs to store large amounts of unstructured data (e.g., images, logs) for training datasets. Which cloud storage solution is purpose-built for data lakes?

Accepted Answer

Amazon S3. Amazon S3 is purpose-built for data lakes because it provides virtually unlimited scalability, high durability (99.999999999% or 11 nines), and supports any type of unstructured data (images, logs, videos) with a flat object storage architecture. Its integration with AWS Glue, Athena, and Lake Formation enables schema-on-read analytics, making it the foundational service for building a data lake on AWS.

Answer

Amazon DynamoDB

Answer

Amazon RDS

Answer

Amazon Redshift

Question 4

A chatbot developer uses a transformer-based model for customer service. Users complain that the chatbot sometimes gives offensive responses. Which technique should be applied first to mitigate this issue?

Accepted Answer

Review and filter the training dataset for offensive or biased language, then fine-tune the model.. The root cause of offensive responses in transformer-based models is typically biased or toxic language present in the training data. Reviewing and filtering the dataset to remove such content, followed by fine-tuning the model, directly addresses the source of the problem. This approach aligns with the principle of data-centric AI, where improving data quality is the first step before modifying model architecture or inference parameters.

Answer

Increase the model size to improve its understanding of context.

Answer

Decrease the temperature parameter to make outputs more deterministic.

Answer

Train a separate classifier to detect offensive outputs in real time.

Question 5

A company trains a large language model on a dataset that includes copyrighted books. Under current legal interpretations, which statement about copyright infringement is MOST accurate?

Accepted Answer

Training on copyrighted data without permission likely infringes copyright, though fair use may be a defense.. Training on copyrighted works without permission is generally considered copyright infringement, unless a specific exception applies (e.g., fair use in the US). Fair use is determined on a case-by-case basis and is not automatically granted. Using only public domain works avoids infringement. The EU AI Act does not provide blanket permission.

Answer

Training on copyrighted data is generally permissible under the EU AI Act.

Answer

Training on copyrighted data is always covered by fair use in the US.

Answer

Training on copyrighted data is allowed as long as the model is not used commercially.

Question 6

While training a deep neural network, the loss function fails to converge and oscillates wildly. Which adjustment is most likely to stabilize training?

Accepted Answer

Reduce the learning rate. When the loss function oscillates wildly and fails to converge, it typically indicates that the learning rate is too high, causing the optimizer to overshoot the minima. Reducing the learning rate allows the gradient descent updates to take smaller, more stable steps, which helps the loss converge smoothly. This is a fundamental hyperparameter tuning step in deep learning training.

Answer

Increase the number of hidden layers

Answer

Decrease the batch size

Answer

Use a test set

Question 7

A company deploys a chatbot using a large language model (LLM). After launch, users report that the chatbot sometimes generates plausible but false information. This phenomenon is known as:

Accepted Answer

Hallucination. Hallucination in LLMs refers to the generation of plausible but factually incorrect or nonsensical information. This occurs when the model's probabilistic next-token prediction produces confident-sounding outputs that deviate from training data or real-world facts, often due to insufficient grounding or training data gaps.

Answer

Gradient explosion

Answer

Overfitting

Answer

Concept drift

Question 8

A data scientist is using differential privacy to protect individual privacy in a training dataset. Which TWO actions are correct implementations of differential privacy?

Accepted Answer

Set a privacy budget (epsilon) to limit information leakage. Setting a privacy budget (epsilon) is a core mechanism in differential privacy that quantifies and limits the amount of information leaked about any individual in the dataset. By controlling epsilon, the data scientist can formally bound the privacy loss, ensuring that the model's outputs do not reveal whether any specific individual's data was included in training.

Answer

Train the model on a small subset of data to reduce exposure

Answer

Remove all personally identifiable information (PII) from the dataset

Answer

Aggregate data into groups before training

Question 9

During training of a neural network, the loss oscillates and does not converge smoothly. The learning rate is set to 0.1. What is the most likely cause and what adjustment should be made?

Accepted Answer

Learning rate too high; decrease it. A learning rate of 0.1 is relatively high for many neural network architectures. When the learning rate is too high, the optimizer takes steps that overshoot the minimum of the loss function, causing the loss to oscillate or even diverge instead of converging smoothly. Decreasing the learning rate allows for smaller, more stable weight updates, leading to smoother convergence.

Answer

Learning rate too low; increase it

Answer

Batch size too small; increase it

Answer

Too many epochs; stop early

Question 10

A data scientist is preparing a dataset for supervised learning. Which TWO steps are essential?

Accepted Answer

Labeling the data. Labeling the data is essential for supervised learning because the algorithm requires input-output pairs to learn a mapping function. Without labeled data, the model cannot be trained to predict outcomes, as supervised learning relies on ground-truth targets for error correction during training.

Answer

One-hot encoding all features

Answer

Normalizing features

Answer

Removing outliers

Question 11

A data science team is preparing a dataset for a binary classification model. The dataset has 95% negative class and 5% positive class. Which technique should they apply to avoid biased model predictions?

Accepted Answer

Apply resampling techniques such as SMOTE or random undersampling. The dataset is severely imbalanced (95% negative vs. 5% positive), which causes classifiers to favor the majority class and produce biased predictions. Resampling techniques such as SMOTE (Synthetic Minority Over-sampling Technique) generate synthetic minority-class samples, while random undersampling reduces majority-class samples, rebalancing the class distribution so the model learns both classes effectively.

Answer

Normalise all numerical features to a [0,1] range

Answer

Shuffle the dataset randomly before splitting into train and test sets

Answer

Remove all rows with missing values

Question 12

Which TWO are best practices for versioning machine learning models? (Choose 2)

Accepted Answer

Tag each model with training date, hyperparameters, and performance metrics. Option B is correct because tagging each model with its training date, hyperparameters, and performance metrics creates an auditable lineage that lets teams reproduce results, compare candidates, and roll back to a known-good model when production metrics degrade. Option C is correct because placing model code and configuration under a version control system such as Git provides immutable commit history, branching, code review, and the ability to correlate a deployed artifact with the exact source revision that produced it. Together, B and C satisfy the core ML versioning requirements of reproducibility, traceability, and governance. Option A is wrong because reusing one model version across all deployments eliminates the ability to distinguish, roll back, or A/B test different models. Option D is wrong because storing only the final binary without metadata makes the model impossible to reproduce, audit, or troubleshoot. Option E is wrong because manually renaming files is error-prone, unauditable, and does not capture training context or enable automated deployment pipelines.

Answer

Use the same model version for all deployments

Answer

Store only the final model binary without metadata

Answer

Manually rename model files with version numbers

Question 13

A company wants to build a system that can generate new product images for an online catalog. Which TWO generative AI approaches are most suitable?

Accepted Answer

Diffusion models. Diffusion models (A) are correct because they generate high-fidelity, photorealistic images by iteratively denoising random noise, making them ideal for producing new product images for an online catalog. Generative Adversarial Networks (C) are also correct because a generator-discriminator pair can synthesize realistic product imagery and can be trained to match a catalog's visual style. Variational autoencoders (B) are not the best fit here because their outputs tend to be blurrier and less photorealistic than diffusion or GAN results. BERT-based models (D) are encoder-only language models for understanding text, not image generation. GPT-style language models (E) generate text, not images, so they do not meet the requirement.

Answer

Variational autoencoders (VAEs)

Answer

BERT-based model

Answer

GPT-style language model

Question 14

A startup is building a chatbot to handle customer inquiries. They want the chatbot to understand context and provide accurate responses without requiring extensive labeled data. Which AI approach is most suitable?

Accepted Answer

Transfer learning with a pre-trained transformer model. Transfer learning with a pre-trained transformer model (e.g., BERT, GPT) is the most suitable approach because it allows the chatbot to understand context and generate accurate responses using knowledge learned from vast general-domain text, requiring only minimal fine-tuning on the startup's specific customer inquiry data. This eliminates the need for extensive labeled datasets, as the model already captures nuanced language patterns and contextual relationships through its self-attention mechanism.

Answer

Reinforcement learning from human feedback

Answer

Rule-based natural language processing

Answer

Convolutional neural networks (CNNs)

Question 15

A company is required to disclose that content has been generated or significantly modified by AI. Which practice directly addresses this transparency obligation?

Accepted Answer

Applying AI watermarking. AI watermarking directly addresses the transparency obligation by embedding a detectable signal into AI-generated content, enabling clear disclosure that the content was produced or significantly modified by AI. This practice aligns with regulatory requirements for provenance and traceability, as watermarks can be verified by automated systems or human inspection to confirm AI origin.

Answer

Using LIME for explanations

Answer

Implementing federated learning

Answer

Publishing a model card