ARTiBA Artificial Intelligence Engineer (AiE®) — Questions and Answers

Question 1: Which trait is frequently connected to artificial intelligence? (a) Consciousness and emotion

Autonomous decision-making without human input
Limited scope and application in specific domains (Correct answer)
Autonomous decision-making without human input
Emotion and consciousness

Correct answer: Limited scope and application in specific domains

Current artificial intelligence systems are primarily examples of 'narrow AI,' meaning they are designed to perform specific tasks within limited domains. Unlike human general intelligence, they lack consciousness, emotion, or the ability to apply knowledge broadly across different contexts. Their intelligence is specialized and confined to their training data and programmed objectives.

Question 2: What is the primary advantage of ensemble methods like Random Forest over a single decision tree?

They eliminate the need for feature engineering
They are easier to interpret and visualize
They reduce variance by aggregating predictions from multiple models (Correct answer)
They require less computational power to train

Correct answer: They reduce variance by aggregating predictions from multiple models

Random Forest reduces variance by averaging predictions from many decision trees trained on random subsets of data and features, resulting in better generalization than a single tree which tends to overfit.

Question 3: An input layer and several sublayers make up the machine in a deep learning network. In order to transport information, the neurons continuously build upon one another. According to the article, "the output of one neuron becomes the input to other neurons in the next layer of the network, and this process continues until the last layer creates the output of the network." A piece of input data is used by the deep learning networks to create something useful.

What is Unsupervised Machine Learning?
What is a Large Language Model (LLM)?
What makes up a Deep Learning Network? (Correct answer)
What are the pros & cons of fine tuning LLMs?

Correct answer: What makes up a Deep Learning Network?

The provided text explicitly details the architectural components of a deep learning network, such as input layers, sublayers, and neurons. It explains how these neurons are interconnected and how information flows from one layer to the next, with the output of one neuron becoming the input for the next. This entire description directly answers the question of what constitutes a deep learning network.

Question 4: What is a vector database's primary role in an LLM application architecture?

Storing and querying high-dimensional embeddings for semantic similarity search (Correct answer)
Storing model weights for fast loading
Caching LLM API responses
Managing prompt templates

Correct answer: Storing and querying high-dimensional embeddings for semantic similarity search

Vector databases (e.g., Pinecone, Weaviate, Chroma) store embeddings and support fast approximate nearest-neighbor search for retrieval in RAG pipelines.

Question 5: What is 'chain-of-thought prompting' in LLMs?

Using multiple models in a pipeline
Prompting the model to reason through intermediate steps before giving a final answer (Correct answer)
Chaining multiple LLM API calls sequentially
Training the model on logical reasoning datasets

Correct answer: Prompting the model to reason through intermediate steps before giving a final answer

Chain-of-thought prompting encourages LLMs to explicitly generate reasoning steps, significantly improving performance on complex reasoning tasks.

Question 6: Which of the following uses artificial intelligence?

Expert Systems
Vision Systems
All of the above (Correct answer)
Gaming

Correct answer: All of the above

Artificial intelligence is a versatile technology applied across numerous fields. It powers intelligent opponents and adaptive gameplay in gaming, forms the core of expert systems for decision support, and enables capabilities like object recognition and facial detection in vision systems. Therefore, AI's utility extends to all the listed applications.

Question 7: When fine-tuning an Azure OpenAI model, what file format is required for the training dataset?

Parquet
JSONL with chat completion format (Correct answer)
XML
CSV with headers

Correct answer: JSONL with chat completion format

Azure OpenAI fine-tuning requires JSONL files where each line is a JSON object with a 'messages' array following the chat completion schema.

Question 8: In k-means clustering, how is the optimal number of clusters (k) typically determined?

By running the algorithm with k=1 and incrementally adding clusters until accuracy stops improving
By using the default value of k=3 as it works for most datasets
By using the elbow method, which plots inertia vs. k and identifies the point of diminishing returns (Correct answer)
By setting k equal to the square root of the number of data points

Correct answer: By using the elbow method, which plots inertia vs. k and identifies the point of diminishing returns

The elbow method plots the within-cluster sum of squares (inertia) against different values of k — the 'elbow' point where adding more clusters yields diminishing reductions in inertia suggests the optimal k.

Question 9: Which evaluation metric is most appropriate when classes are severely imbalanced and the cost of false negatives is high?

Accuracy
F1 Score (Correct answer)
R-squared
Mean Squared Error

Correct answer: F1 Score

The F1 Score is the harmonic mean of precision and recall, making it suitable for imbalanced datasets where accuracy is misleading — it balances the cost of false positives and false negatives.

Question 10: Which principle of responsible AI ensures that stakeholders can understand and interrogate how an AI system makes decisions?

Data compression
Throughput optimization
Scalability
Explainability and transparency (Correct answer)

Correct answer: Explainability and transparency

Explainability means AI decisions can be understood by humans, while transparency ensures the system's design, data, and limitations are disclosed.

Question 11: What does 'consent' mean in the context of using personal data to train AI models?

Getting approval from the AI ethics board
Having executives approve the training dataset
Obtaining informed agreement from individuals before using their personal data for model training (Correct answer)
Signing an NDA with data providers

Correct answer: Obtaining informed agreement from individuals before using their personal data for model training

Consent requires that individuals knowingly agree to how their personal data will be used, including for AI training, per regulations like GDPR and CCPA.

Question 12: What is RLHF (Reinforcement Learning from Human Feedback) used for in LLM development?

Speeding up pretraining
Expanding the model's vocabulary
Aligning LLM outputs with human preferences and reducing harmful outputs (Correct answer)
Compressing the model for edge deployment

Correct answer: Aligning LLM outputs with human preferences and reducing harmful outputs

RLHF uses human preference ratings to train a reward model, then fine-tunes the LLM with RL to produce outputs humans rate as better and safer.

Question 13: A product catalog can be searched using the Bing Search service you have. You must locate the following details: - The locale of the query - The top 50 query strings - The number of calls to the service - The top geographical regions of the service What should you implement?

Azure API Management (APIM)
Azure Application Insights
Bing Statistics (Correct answer)
Azure Monitor

Correct answer: Bing Statistics

Bing Statistics (or Bing Webmaster Tools analytics, which includes similar data) provides detailed insights into how users are interacting with Bing Search services. It offers metrics such as query locale, popular search queries, total calls to the service, and geographical distribution of users. This makes it the direct and most appropriate tool for gathering the specified usage and performance data for a Bing Search service.

Question 14: What does Principal Component Analysis (PCA) accomplish?

It generates synthetic training samples to balance imbalanced datasets
It selects the most important features based on their correlation with the target variable
It reduces dimensionality by projecting data onto orthogonal axes of maximum variance (Correct answer)
It clusters data points into groups based on their proximity in feature space

Correct answer: It reduces dimensionality by projecting data onto orthogonal axes of maximum variance

PCA finds orthogonal principal components (linear combinations of original features) ordered by the amount of variance they explain, allowing dimensionality reduction by keeping only the top components.

Question 15: What is 'embedding' in NLP, as used by transformer models?

The process of tokenizing input text
A dense, fixed-size vector representation of a token or text that captures semantic meaning (Correct answer)
Storing model weights in a database
A technique for compressing the attention matrix

Correct answer: A dense, fixed-size vector representation of a token or text that captures semantic meaning

Embeddings map discrete tokens (or entire texts) to dense vectors in a continuous space where semantically similar items are geometrically close.

Question 16: What is the primary challenge addressed by 'federated learning' in AI?

Training AI models across decentralized devices without sharing raw data, preserving privacy (Correct answer)
Training models across multiple cloud providers
Distributing training compute across multiple GPUs
Federating API access to AI models

Correct answer: Training AI models across decentralized devices without sharing raw data, preserving privacy

Federated learning trains models locally on each device and aggregates only model updates (not raw data), enabling collaboration without centralizing sensitive data.

Question 17: Which of the following statements best describes a naive Bayes classifier?

It iteratively adjusts weights to minimize the cross-entropy loss between predictions and labels
It applies Bayes' theorem assuming conditional independence between features given the class label (Correct answer)
It partitions the feature space into rectangular regions using a series of threshold decisions
It builds a decision boundary by maximizing the margin between classes

Correct answer: It applies Bayes' theorem assuming conditional independence between features given the class label

Naive Bayes applies Bayes' theorem to compute the posterior probability of each class and classifies based on the highest probability, making the 'naive' assumption that features are conditionally independent given the class.

Question 18: An example of a production rule is .

A sequence of steps
Arbitrary representation to problem
A set of Rule
Set of Rule & Sequence of steps (Correct answer)

Correct answer: Set of Rule & Sequence of steps

A production rule in AI is a fundamental building block for knowledge representation and reasoning, typically structured as an 'IF-THEN' statement. These rules, when combined with a sequence of steps or an inference engine, allow an AI system to make decisions, draw conclusions, or perform actions based on specific conditions. Thus, it involves both a set of rules and their sequential application.

Question 19: What is the process of adjusting a machine learning model's parameters to minimize errors on the training data?

Overfitting
Underfitting
Optimization (Correct answer)
Regression

Correct answer: Optimization

Optimization is the process of iteratively adjusting a machine learning model's internal parameters (like weights and biases) to minimize a defined error or loss function. This process aims to find the best set of parameters that allows the model to make the most accurate predictions on the training data. Algorithms like gradient descent are commonly used for this purpose.

Question 20: What is the context window in a large language model?

The time window used for model training
The maximum number of tokens the model can process in a single input/output sequence (Correct answer)
The layer of attention that focuses on the current word
The UI window showing model outputs

Correct answer: The maximum number of tokens the model can process in a single input/output sequence

The context window defines the maximum number of tokens (input + output) an LLM can process at once, limiting how much text it can consider.

Question 21: Which evaluation metric measures the overlap between generated text and human reference text using n-gram precision?

ROUGE
Perplexity
BLEU (Correct answer)
F1 Score

Correct answer: BLEU

BLEU (Bilingual Evaluation Understudy) measures n-gram precision between generated and reference text, commonly used for machine translation and text generation evaluation.

Question 22: Which US federal principle guides that AI systems should behave reliably and safely across a range of conditions?

Reliability and safety (Correct answer)
Accountability
Inclusiveness
Privacy and security

Correct answer: Reliability and safety

Reliability and safety is a core responsible AI principle requiring systems to perform as intended under varied conditions without unintended harmful outcomes.

Question 23: What is 'algorithmic bias' in AI systems?

High variance in model performance across runs
Overfitting to training data
Systematic and unfair discrimination in AI outputs caused by biased training data or model design (Correct answer)
Random errors in model predictions

Correct answer: Systematic and unfair discrimination in AI outputs caused by biased training data or model design

Algorithmic bias occurs when an AI system produces systematically unfair outcomes for certain groups, often reflecting historical biases in training data.

Question 24: What is 'zero-shot prompting' when using an LLM?

Using a model that has never been fine-tuned
Asking the model to perform a task without providing any examples in the prompt (Correct answer)
Prompting the model with zero tokens
Training the model on zero labeled examples

Correct answer: Asking the model to perform a task without providing any examples in the prompt

Zero-shot prompting asks the LLM to complete a task using only instructions, relying entirely on knowledge from pretraining without in-context examples.

Question 25: What is 'AI red-teaming'?

Training AI models on red-flagged data
A technique for reducing LLM inference cost
A competitive AI engineering tournament
Adversarial testing where experts try to find failures, vulnerabilities, and harmful behaviors in AI systems (Correct answer)

Correct answer: Adversarial testing where experts try to find failures, vulnerabilities, and harmful behaviors in AI systems

AI red-teaming involves deliberately probing a system for safety failures, harmful outputs, or exploitable behaviors before deployment, mimicking adversarial users.

Question 26: Which metric measures how well an LLM predicts a test corpus, with lower values indicating better language modeling?

AUC-ROC
BLEU
Precision
Perplexity (Correct answer)

Correct answer: Perplexity

Perplexity measures the exponentiated average negative log-likelihood of a test set; lower perplexity means the model assigns higher probability to the observed text.

Question 27: What is the 'attention mechanism' in transformer-based models?

A regularization method for language models
A technique for data augmentation in NLP
A mechanism that computes weighted relationships between all positions in a sequence (Correct answer)
A method to prune unimportant neurons

Correct answer: A mechanism that computes weighted relationships between all positions in a sequence

Attention computes a weighted sum of values based on query-key similarity, allowing the model to focus on relevant parts of the input regardless of distance.

Question 28: What is the curse of dimensionality and how does it affect machine learning models?

High-dimensional data requires exponentially more training examples to maintain statistical coverage of the feature space (Correct answer)
Deep learning models cannot process inputs with more than 1,000 features
Models with many hyperparameters always overfit regardless of training data size
Adding more features always improves model performance due to increased information

Correct answer: High-dimensional data requires exponentially more training examples to maintain statistical coverage of the feature space

As the number of dimensions increases, the volume of the feature space grows exponentially, making data increasingly sparse — meaning far more training examples are needed to learn reliable patterns.

Question 29: What is 'semantic chunking' in the context of building RAG systems?

Splitting documents by fixed token counts
Encrypting document chunks before embedding
Filtering out low-quality document segments
Dividing documents into chunks based on semantic coherence to preserve meaningful context (Correct answer)

Correct answer: Dividing documents into chunks based on semantic coherence to preserve meaningful context

Semantic chunking splits documents at natural semantic boundaries rather than fixed sizes, improving retrieval quality by keeping coherent content together.

Question 30: What is the purpose of a validation set, distinct from both the training set and test set?

To tune hyperparameters and select the best model without contaminating the final test evaluation (Correct answer)
To provide additional data for training when the dataset is small
To evaluate the final model performance and report results to stakeholders
To detect data drift after a model is deployed to production

Correct answer: To tune hyperparameters and select the best model without contaminating the final test evaluation

The validation set is used during development to tune hyperparameters and compare models — using the test set for this purpose would cause leakage, making the test set an unreliable measure of real-world performance.

Question 31: What is early stopping as a regularization technique during model training?

Halting training when validation loss stops improving to prevent overfitting (Correct answer)
Removing neurons with the lowest activation values after each epoch
Truncating the gradient computation at a maximum value to prevent exploding gradients
Reducing the learning rate exponentially after a fixed number of epochs

Correct answer: Halting training when validation loss stops improving to prevent overfitting

Early stopping monitors validation loss during training and halts the process when it starts increasing, saving the model weights from the epoch with the best validation performance to prevent overfitting.

Question 32: What is 'data governance' in the context of AI system development?

Governing GPU resource allocation during training
A framework of policies and processes for ensuring data quality, privacy, security, and compliance throughout the AI lifecycle (Correct answer)
Managing version control for training scripts
Controlling who can deploy AI models

Correct answer: A framework of policies and processes for ensuring data quality, privacy, security, and compliance throughout the AI lifecycle

Data governance defines who owns data, how it's collected and used, retention policies, and ensures compliance with privacy regulations throughout the AI pipeline.

Question 33: Which approach is most effective for detecting and reducing 'factual drift' in long-context language model outputs?

Using greedy decoding exclusively to avoid sampling variance
Reducing the context window to force shorter responses
Increasing the model's temperature setting
Applying self-consistency checks or chain-of-thought verification against source documents (Correct answer)

Correct answer: Applying self-consistency checks or chain-of-thought verification against source documents

Self-consistency methods and chain-of-thought prompting that explicitly cites source passages help detect when generated claims diverge from grounded facts as output length grows.

Question 34: Which technique reduces LLM hallucinations by grounding responses in retrieved documents?

Fine-tuning on curated datasets
Increasing model temperature
Chain-of-thought prompting
Retrieval-Augmented Generation (RAG) (Correct answer)

Correct answer: Retrieval-Augmented Generation (RAG)

RAG retrieves relevant documents at inference time and conditions the model's output on that grounded context, reducing hallucinations.

Question 35: In gradient descent, what does the learning rate control?

The proportion of data used in each training batch
The size of the steps taken toward the minimum of the loss function (Correct answer)
The threshold for classifying a prediction as positive
The number of training epochs before the model converges

Correct answer: The size of the steps taken toward the minimum of the loss function

The learning rate determines how large each update step is during gradient descent — too large causes oscillation or divergence, while too small results in slow convergence or getting stuck in local minima.

Question 36: What is 'multi-hop reasoning' in knowledge graphs and QA systems?

Running the same query multiple times for consistency checks
Splitting long documents into multiple chunks before indexing
Answering a question by chaining multiple inference steps across graph edges (Correct answer)
Parallelizing knowledge retrieval across multiple servers

Correct answer: Answering a question by chaining multiple inference steps across graph edges

Multi-hop reasoning requires traversing multiple relationships in a knowledge graph or making several logical inferences to arrive at an answer that cannot be found in a single step.

Question 37: What is Retrieval-Augmented Generation (RAG)?

Combining an LLM with a retrieval system to ground responses in external documents (Correct answer)
A technique for compressing LLM context windows
A method for training LLMs on retrieval tasks
Using multiple LLMs in an ensemble

Correct answer: Combining an LLM with a retrieval system to ground responses in external documents

RAG retrieves relevant documents from an external knowledge base and injects them into the LLM's context, enabling up-to-date, grounded responses.

Question 38: A developer wants to detect whether a customer's message expresses frustration. Which Azure AI Language feature is most appropriate?

Text summarization
Language detection
Custom text classification
Sentiment analysis with opinion mining (Correct answer)

Correct answer: Sentiment analysis with opinion mining

Sentiment analysis with opinion mining evaluates the emotional tone of text and can identify negative sentiment indicative of frustration.

Question 39: What is 'model cards' documentation in responsible AI?

Structured documents reporting a model's intended uses, performance across subgroups, limitations, and ethical considerations (Correct answer)
API documentation for model endpoints
Business cards for AI engineers
Training configuration files

Correct answer: Structured documents reporting a model's intended uses, performance across subgroups, limitations, and ethical considerations

Model cards, introduced by Google, standardize transparency by documenting a model's purpose, performance benchmarks, fairness evaluations, and known limitations.

Question 40: Which of the following best describes overfitting in a machine learning model?

The model converges before reaching the global minimum of the loss function
The model requires too many features to make accurate predictions
The model performs poorly on both training and test data
The model performs well on training data but poorly on unseen test data (Correct answer)

Correct answer: The model performs well on training data but poorly on unseen test data

Overfitting occurs when a model learns the noise and patterns specific to training data so well that it fails to generalize — resulting in high training accuracy but poor test accuracy.

Question 41: What is the main benefit of using a directed acyclic graph (DAG) to represent an ML pipeline?

It compresses model weights
It increases model accuracy
It explicitly defines task dependencies, enabling parallel execution and reproducibility (Correct answer)
It reduces training data size

Correct answer: It explicitly defines task dependencies, enabling parallel execution and reproducibility

A DAG maps task dependencies so the orchestrator can parallelize independent steps and ensure correct execution order for reproducible pipelines.

Question 42: Why is feature standardization (zero mean, unit variance) important before applying algorithms like SVM or k-nearest neighbors?

It prevents features with larger numerical ranges from dominating distance-based calculations (Correct answer)
It automatically removes irrelevant features from the dataset
It ensures the model produces probabilistic outputs between 0 and 1
It reduces the number of training iterations required for convergence

Correct answer: It prevents features with larger numerical ranges from dominating distance-based calculations

Distance-based algorithms are sensitive to feature scale — a feature ranging 0–10,000 will dominate a feature ranging 0–1 in distance calculations, so standardization puts all features on equal footing.

Question 43: What is 'hallucination' in the context of LLMs?

The model generating extremely long outputs
The model refusing to answer sensitive questions
The model outputting garbled text due to tokenization errors
The model producing plausible-sounding but factually incorrect or fabricated information (Correct answer)

Correct answer: The model producing plausible-sounding but factually incorrect or fabricated information

LLM hallucination refers to the model confidently generating false, invented information not grounded in training data or retrieved context.

Question 44: What is an 'AI incident' as defined in responsible AI frameworks?

A drop in model accuracy below a threshold
An event where an AI system causes or contributes to harm, near-miss, or unexpected negative consequences in deployment (Correct answer)
A model failing to converge during training
A disagreement between AI engineers about model architecture

Correct answer: An event where an AI system causes or contributes to harm, near-miss, or unexpected negative consequences in deployment

An AI incident is any real-world event where an AI system causes harm or poses significant risk, tracked in repositories like the AI Incident Database.

Question 45: What is 'differential privacy' in the context of AI and data science?

A technique for removing outliers from training data
Encrypting model weights
A mathematical framework that adds noise to data or computations to protect individual privacy while enabling aggregate analysis (Correct answer)
A method to compare two datasets

Correct answer: A mathematical framework that adds noise to data or computations to protect individual privacy while enabling aggregate analysis

Differential privacy provides mathematical guarantees that an individual's data cannot be inferred from model outputs by injecting carefully calibrated noise.

Question 46: What does 'temperature' control in LLM text generation?

The randomness of token sampling — higher values produce more diverse outputs (Correct answer)
The computational load during inference
The model's confidence threshold for answering
The maximum length of generated text

Correct answer: The randomness of token sampling — higher values produce more diverse outputs

Temperature scales the logits before softmax; higher values flatten the distribution (more random), lower values sharpen it (more deterministic).

Question 47: What is one possible ethical issue connected to the usage of artificial intelligence?

Overreliance on human judgment and decision-making
Potential for biased or discriminatory outcomes (Correct answer)
Advancement in scientific research and discoveries
Decreased need for data privacy and security measures

Correct answer: Potential for biased or discriminatory outcomes

One significant ethical issue with AI is the potential for algorithms to perpetuate or even amplify existing societal biases present in the data they are trained on. This can lead to discriminatory outcomes in areas like hiring, lending, or criminal justice, impacting fairness and equity. Addressing bias in AI development is a critical concern.

Question 48: Which activation function is most commonly used in hidden layers of modern deep neural networks to mitigate the vanishing gradient problem?

ReLU (Rectified Linear Unit) (Correct answer)
Sigmoid
Softmax
Tanh

Correct answer: ReLU (Rectified Linear Unit)

ReLU outputs max(0, x), maintaining gradient magnitude for positive inputs and largely avoiding vanishing gradients compared to sigmoid or tanh.

Question 49: What is the primary purpose of k-fold cross-validation?

To obtain a more reliable estimate of model performance by using all data for both training and validation (Correct answer)
To select the optimal number of features for a model
To reduce training time by splitting data into smaller batches
To increase the size of the training dataset through data augmentation

Correct answer: To obtain a more reliable estimate of model performance by using all data for both training and validation

K-fold cross-validation splits data into k subsets, trains on k-1 folds, and validates on the remaining fold, rotating until all folds are used — giving a robust performance estimate without wasting data.

Question 50: What is 'vector similarity search' used for in LLM-powered applications?

Detecting prompt injection attacks
Finding the most semantically similar documents to a query by comparing embedding vectors (Correct answer)
Training transformer attention weights
Reducing LLM inference latency

Correct answer: Finding the most semantically similar documents to a query by comparing embedding vectors

Vector similarity search retrieves documents whose embeddings are closest to a query embedding, powering RAG, semantic search, and recommendation systems.

Question 51: What does 'scalable oversight' aim to solve in AI alignment research?

Scaling model parameters beyond 1 trillion
Automating model retraining at scale
Distributing AI workloads across clusters
Maintaining meaningful human supervision of AI as systems become more capable than the humans overseeing them (Correct answer)

Correct answer: Maintaining meaningful human supervision of AI as systems become more capable than the humans overseeing them

Scalable oversight addresses how to keep humans in meaningful control of AI decisions when the AI may eventually be more capable than the humans evaluating it.

Question 52: What does 'tokenization' mean in the context of NLP preprocessing?

Encrypting text data before storage
Removing stopwords from a document
Converting text to audio
Splitting raw text into discrete units (tokens) for model input (Correct answer)

Correct answer: Splitting raw text into discrete units (tokens) for model input

Tokenization breaks raw text into tokens (words, subwords, or characters) that are then mapped to numeric IDs for model processing.

Question 53: What is the primary difference between a parametric and a non-parametric machine learning model?

Parametric models can only handle continuous features, while non-parametric models handle categorical features
Parametric models require GPU acceleration, while non-parametric models run on CPUs
Parametric models use gradient descent, while non-parametric models use Bayesian inference
Parametric models assume a fixed functional form with a set number of parameters, while non-parametric models grow in complexity with the training data (Correct answer)

Correct answer: Parametric models assume a fixed functional form with a set number of parameters, while non-parametric models grow in complexity with the training data

Parametric models (e.g., linear regression, logistic regression) summarize data with a fixed number of parameters regardless of dataset size, while non-parametric models (e.g., k-NN, kernel SVM) retain training data and grow in complexity with more data.

Question 54: What is 'knowledge distillation' in the context of AI model compression?

Removing duplicate facts from a knowledge base
Training a smaller student model to mimic a larger teacher model (Correct answer)
Extracting rules from a neural network's weights
Converting symbolic knowledge to vector embeddings

Correct answer: Training a smaller student model to mimic a larger teacher model

Knowledge distillation transfers knowledge from a large, complex teacher model to a smaller student model by training the student on soft probability outputs from the teacher.

Question 55: Which search technique uses the least memory?

Linear Search
Breadth-First search
Depth-First Search (Correct answer)
Optimal search

Correct answer: Depth-First Search

Depth-First Search (DFS) uses the least memory among common graph traversal algorithms because it only needs to store the current path from the root to the current node. It explores as far as possible along each branch before backtracking, making its space complexity proportional to the depth of the search tree, which is generally less than Breadth-First Search for deep trees.

Question 56: What does the ROC-AUC score measure in a binary classification model?

The probability that the model ranks a randomly chosen positive example higher than a randomly chosen negative example (Correct answer)
The average squared difference between predicted probabilities and true labels
The ratio of true positives to the total number of actual positive examples
The proportion of predictions that match the true class label across all thresholds

Correct answer: The probability that the model ranks a randomly chosen positive example higher than a randomly chosen negative example

AUC (Area Under the ROC Curve) represents the probability that the model assigns a higher predicted probability to a random positive instance than to a random negative instance — an AUC of 1.0 is perfect, 0.5 is random.

Question 57: In contrast to machine learning (ML), which of the following is a potential result of artificial intelligence (AI)?

AI can autonomously learn and adapt to new situations, while ML relies on pre-defined algorithms. (Correct answer)
AI is more suitable for handling complex and unstructured data than ML.
ML has a broader scope and can perform a wider range of tasks than AI.
ML requires more computational power and resources compared to AI.

Correct answer: AI can autonomously learn and adapt to new situations, while ML relies on pre-defined algorithms.

While machine learning (ML) is a subset of artificial intelligence (AI) that focuses on systems learning from data, AI encompasses a broader goal of creating intelligent agents that can reason, problem-solve, and adapt. Advanced AI systems can exhibit autonomous learning and adaptation to new situations, going beyond the pre-defined algorithms that typically characterize ML models. This allows AI to handle more complex, dynamic environments.

Question 58: What is the primary difference between gradient boosting and bagging (e.g., Random Forest)?

Bagging builds models sequentially, while gradient boosting builds them in parallel
Gradient boosting averages predictions, while bagging uses voting to select the best model
Gradient boosting builds models sequentially where each model corrects the errors of the previous one (Correct answer)
Bagging is only applicable to regression tasks, while gradient boosting handles classification

Correct answer: Gradient boosting builds models sequentially where each model corrects the errors of the previous one

Gradient boosting builds an ensemble sequentially, where each new model focuses on correcting the residual errors of the combined previous models, reducing bias — unlike bagging which builds models independently in parallel to reduce variance.

Question 59: Which Azure OpenAI parameter controls how deterministic or random the model's output is during text generation?

top_p
temperature (Correct answer)
frequency_penalty
max_tokens

Correct answer: temperature

Temperature scales the probability distribution over tokens; lower values make output more deterministic, higher values increase randomness.

Question 60: In AI system design, what is the purpose of a 'human-in-the-loop' mechanism?

To replace AI with human workers
To monitor server infrastructure
To include human oversight and intervention in AI decision pipelines for critical or uncertain cases (Correct answer)
To train models using human-labeled data

Correct answer: To include human oversight and intervention in AI decision pipelines for critical or uncertain cases

Human-in-the-loop keeps humans involved in reviewing, correcting, or approving AI decisions, especially where errors have high stakes.

ARTiBA Artificial Intelligence Engineer (AiE®)

The AiE® certification validates expertise across core AI engineering domains including machine learning, neural networks, NLP, and responsible AI deployment, aligned with the AMDEX™ Knowledge Framework. It is a globally recognized credential for professionals building and implementing AI systems.

Questions

90 min

Time Limit

70%

To Pass

Checking for saved progress...

Exam Rules

You can skip questions and return to them later
Flag questions for review before submitting
No feedback shown until you submit the entire exam
Unanswered questions count as wrong — answer everything
10 pretest questions are mixed in and don't affect your score
Timer auto-submits when time runs out
Your progress is auto-saved every 30 seconds