AI0-001 Practice Test 46 — 15 Questions

Question 1

A data scientist is preparing a dataset for training a customer churn prediction model. To prevent train/test leakage, which TWO practices should be followed? (Select TWO)

Accepted Answer

Split the data chronologically (e.g., use data before a certain date for training, after for testing). Option C is correct because splitting data chronologically (e.g., training on records before a cutoff date and testing on records after that date) respects the temporal order of observations and prevents future information from leaking into the training set, which is essential for time-dependent churn prediction. Option E is correct because feature selection must be performed using only the training data; if the test set influences which features are selected, information from the test set leaks into model development and produces overly optimistic performance estimates. Option A is incorrect because removing duplicates only from the test set does not prevent leakage and can distort the test distribution; duplicate handling should be consistent and decided before splitting. Option B is incorrect because random shuffling of the entire dataset before splitting can mix past and future observations, which is especially harmful for temporal churn data and does not by itself prevent leakage. Option D is incorrect because computing normalization statistics on the entire dataset before splitting leaks test-set distribution information into training; normalization statistics must be computed only on the training set and then applied to the test set.

Answer

Remove duplicate records only from the test set to ensure uniqueness

Answer

Shuffle the entire dataset randomly before splitting into train and test sets

Answer

Normalize numerical features using statistics computed on the entire dataset before splitting

Question 2

An organization wants to train a machine learning model on sensitive patient data without exposing individual records. Which privacy-preserving technique allows the model to learn from data distributed across multiple hospitals without raw data leaving each site?

Accepted Answer

Federated learning. Federated learning trains a shared global model by sending model updates (gradients or weights) — not raw data — from each participating site to a central aggregator, which combines them (e.g., via FedAvg) and redistributes the updated model. Because patient records never leave each hospital, it directly satisfies the requirement of learning from distributed data without exposing individual records. This is the canonical privacy-preserving distributed training technique.

Answer

Homomorphic encryption

Answer

k-anonymity

Answer

Differential privacy

Question 3

An AI model for skin cancer detection achieves high accuracy but performs poorly on dark skin tones. The team wants to evaluate whether the model is calibrated across skin tones. Which fairness metric should they use?

Accepted Answer

Calibration. Calibration is the correct metric because it directly measures whether the predicted probabilities of skin cancer match the actual outcomes across different skin tones. A model can have high overall accuracy but be miscalibrated for a subgroup if its confidence scores are systematically over- or under-confident for that group. In this scenario, the team needs to check if the model's risk scores are equally reliable for dark skin tones as for light skin tones, which is exactly what calibration assesses.

Answer

Equalised odds

Answer

Demographic parity

Answer

Individual fairness

Question 4

A team is building a recommendation system for an e-commerce platform. They need to update recommendations in real-time as users browse. Which integration pattern is MOST suitable?

Accepted Answer

AI microservices with a REST API. AI microservices with a REST API allow the recommendation engine to be decomposed into independent, scalable services that can be invoked synchronously on each user interaction, returning fresh recommendations in real time. REST provides a lightweight, stateless request/response contract that fits low-latency, per-request inference, and microservices let the recommendation model scale horizontally and be updated independently of the rest of the e-commerce platform. This combination directly satisfies the 'real-time as users browse' requirement without coupling the model to a monolith or introducing batch/async delays.

Answer

Streaming responses from a single monolithic model

Answer

Batch processing with nightly updates

Answer

Async processing queue with delayed responses

Question 5

A company develops an internal LLM-based tool that queries a vector database containing confidential customer data. Which security measure should be implemented to prevent the LLM from revealing sensitive information in its responses?

Accepted Answer

Output filtering with regex and moderation classifiers. Output filtering with regex and moderation classifiers (Option D) is the correct security measure because it directly inspects the LLM's generated responses for sensitive data patterns (e.g., credit card numbers, PII) and blocks or redacts them before delivery. This prevents the LLM from inadvertently leaking confidential customer data retrieved from the vector database, even if the model's training or prompt injection causes it to include such information in its output.

Answer

Rate limiting on API calls

Answer

Input validation and sanitization

Answer

Audit logging of AI interactions

Question 6

A team is implementing a RAG system for a legal document Q&A. They need to chunk documents effectively. Which THREE chunking strategies should they consider to improve retrieval accuracy for legal texts that contain hierarchical sections (clauses, sub-clauses, definitions)?

Accepted Answer

Hierarchical chunking that indexes chunks at clause and sub-clause levels with parent relationships. Option A is correct because hierarchical chunking indexes chunks at clause and sub-clause levels while preserving parent relationships, which matches the legal document's nested structure and lets retrieval return a precise sub-clause while still offering the enclosing clause for context. Option D is correct because chunking based on the document's table of contents and section hierarchy aligns chunk boundaries with the document's own logical divisions, so each chunk corresponds to a meaningful legal section rather than an arbitrary token span. Option E is correct because semantic chunking splits at natural boundaries such as section headings and paragraph breaks, keeping definitions and clauses intact and avoiding mid-sentence or mid-clause cuts that would degrade embedding quality and retrieval accuracy. Option B is not among the correct answers because a fixed 10% overlap is a generic heuristic that does not respect legal hierarchy and can duplicate or fragment clause text. Option C is not among the correct answers because fixed-size 512-token windows with no overlap ignore section structure and will frequently split clauses, sub-clauses, and definitions across chunk boundaries.

Answer

Overlapping chunks with a 10% overlap between consecutive chunks

Answer

Fixed-size chunking with a 512-token window and no overlap

Question 7

Which similarity search metric is BEST for comparing dense vector embeddings when the magnitude of the vectors is not important, only the direction?

Accepted Answer

Cosine similarity. Cosine similarity measures the cosine of the angle between two vectors, focusing only on their direction and ignoring magnitude. It is ideal for comparing dense vector embeddings when the magnitude is not important, as it normalizes the vectors. This makes it the best choice for tasks like text similarity where the length of the document should not affect the similarity score.

Answer

Euclidean distance

Answer

Dot product

Answer

Manhattan distance

Question 8

A data engineer needs to design a data pipeline for a real-time fraud detection system. The system requires low-latency processing of streaming transactions. Which architecture is most appropriate?

Accepted Answer

Stream processing with Apache Kafka and Flink. Apache Kafka provides a distributed, fault-tolerant event streaming platform that ingests high-throughput transaction data with low latency, while Apache Flink offers true stream processing with exactly-once semantics and sub-second event-time processing. Together, they enable real-time fraud detection by analyzing transactions as they arrive, without the delays inherent in batch or micro-batch approaches.

Answer

Data lake with Apache Spark

Answer

Batch processing with Apache Hadoop

Answer

Microservices architecture with REST APIs

Question 9

During inference, a model served via a REST API occasionally returns high latency due to cold starts. The team uses a containerized service on Kubernetes with horizontal pod autoscaling. Which solution minimizes cold start impact while controlling cost?

Accepted Answer

Use a sidecar container that pre-warms the model and set a minimum replica count. A sidecar warm-up agent and a minimum replica count keep pods ready. Increasing resources may not fix cold starts; autoscaling based on request count may lag; vertical scaling helps but not directly.

Answer

Configure the autoscaler based on request count with a shorter cooldown period

Answer

Increase CPU and memory requests for the inference container

Answer

Switch to vertical pod autoscaling

Question 10

A data engineering team is designing a data pipeline to process streaming sensor data and feed it into an ML model for anomaly detection. Which THREE components are essential for this pipeline?

Accepted Answer

Amazon S3 as a data lake for storing raw sensor data. Amazon S3 is essential as a data lake for storing raw sensor data because it provides durable, scalable, and cost-effective object storage that can serve as a central repository for streaming data before and after processing. In a streaming pipeline, raw data must be persisted for reprocessing, historical analysis, and compliance, and S3's integration with Apache Spark and Kafka makes it a natural landing zone for sensor data.

Answer

Apache Airflow for scheduling recurring batch jobs

Answer

Snowflake as a real-time streaming destination

Question 11

Which TWO of the following are common activation functions used in neural networks? (Choose two.)

Accepted Answer

ReLU. ReLU (Rectified Linear Unit) is a widely used activation function that outputs the input directly if it is positive, and zero otherwise, introducing non-linearity while mitigating the vanishing gradient problem. Sigmoid is another common activation function that maps any real-valued input to a value between 0 and 1, making it useful for binary classification output layers. Both are fundamental building blocks in neural network architectures.

Answer

Gradient descent

Answer

LSTM

Answer

Dropout

Question 12

An AI operations team is monitoring a deployed image classification model. They notice a gradual increase in prediction confidence but a drop in accuracy. Which THREE actions should they take to diagnose the issue?

Accepted Answer

Analyze the model's calibration curve to see if confidence scores align with actual accuracy.. Option A is correct because a calibration curve (reliability diagram) directly compares predicted confidence against observed accuracy, revealing whether the model has become overconfident — exactly the symptom of rising confidence with falling accuracy. Option C is correct because comparing training versus production input feature distributions (e.g., via drift metrics like PSI or KL divergence) detects data drift or covariate shift that can degrade accuracy while inflating confidence. Option D is correct because evaluating on a held-out test set collected at deployment time establishes a stable baseline, distinguishing genuine model degradation from changes in the production data distribution. Option B is not appropriate because collecting unlabeled data does not diagnose the cause and cannot be used for supervised evaluation without labels. Option E is not appropriate because immediate retraining without root-cause analysis risks masking the problem and may reinforce the drift or labeling issues causing the confidence-accuracy mismatch.

Answer

Increase the size of the training dataset by collecting more unlabeled data.

Answer

Retrain the model immediately with the most recent data.

Question 13

An organization wants to implement an AI system to automatically categorize support tickets into predefined categories. They have a labeled dataset of 10,000 tickets. Which approach is MOST appropriate?

Accepted Answer

Fine-tune a pre-trained text classification model. Fine-tuning a pre-trained text classification model is a standard and effective approach for supervised classification when labeled data is available.

Answer

Use a rule-based system with keyword matching

Answer

Use a prompt-based LLM with few-shot examples

Answer

Train a custom neural network from scratch

Question 14

An AI model is being developed for medical diagnosis from X-ray images. The dataset contains only frontal chest X-rays. The model achieves high accuracy on test set but fails on lateral views. What is the most likely cause?

Accepted Answer

Dataset bias. The model was trained exclusively on frontal chest X-rays, so it never learned features specific to lateral views. When tested on lateral views, the distribution shift causes poor performance, which is a classic case of dataset bias (sampling bias). The high accuracy on the test set is misleading because the test set also only contained frontal views, masking the model's inability to generalize to other X-ray orientations.

Answer

Underfitting

Answer

Label noise

Answer

Overfitting

Question 15

A machine learning engineer needs to choose an algorithm for grouping customers into segments based on purchasing behavior without any labels. Which algorithm should the engineer use?

Accepted Answer

K-means clustering. K-means clustering is an unsupervised learning algorithm that groups unlabeled data into clusters based on feature similarity, making it ideal for segmenting customers by purchasing behavior without predefined labels. It partitions data into K clusters by minimizing within-cluster variance, which directly addresses the requirement of discovering natural groupings in the data.

Answer

Random forest classifier

Answer

Linear regression

Answer

Support vector machine