AWS AI Practitioner Fundamentals of AI and ML: 223 practice questions

12 of the 223 Fundamentals of AI and ML questions in the Certsqill AWS AI Practitioner bank, shown in full below. Each one carries an explanation for every option, not just the correct one — the wrong answers are where the marks go.

Preparing for AWS AI Practitioner? Take the free 5-min readiness check →

1. Use machine learning to predict discontinuation: Which approach best meets this requirement?

1. Use a rules engine with conditions defined by education specialists.

   Rules apply predefined logic and do not learn predictive relationships from historical outcomes.
2. Use deep learning to model the outcome with multiple neural-network layers.

   Deep learning can support prediction, but the requirement does not indicate that multilayer neural networks are necessary.
3. Use machine learning to predict discontinuation from historical student data. ✓

   Machine learning learns relationships from labeled historical examples and applies them to new student records.
4. Use a generative model to write personalized explanations of likely discontinuation.

   Generating explanations is different from selecting or predicting which students may discontinue.

Machine learning fits because it learns predictive patterns from historical examples.

2. Deep learning: Which technology is most directly associated with this requirement?

1. Unsupervised clustering of travel bookings

   Clustering groups similar records but does not directly identify named landmarks in uploaded photographs.
2. Rule-based matching with fixed landmark conditions

   Fixed rules are difficult to maintain for varied photographs, viewpoints, lighting, and landmark appearances.
3. Deep learning ✓

   Deep learning uses multilayer neural networks and is commonly suited to complex image-recognition patterns.
4. Generative text completion for itinerary descriptions

   Text generation can create descriptions but does not itself provide landmark image recognition.

Deep learning uses multilayer neural networks and is well suited to recognizing complex patterns in images.

3. Generative output: Which output type best matches this requirement?

1. Unsupervised grouping output

   Clustering groups similar products without supplied labels but does not generate descriptive text.
2. Generative output ✓

   Generative AI creates new content, such as natural-language descriptions, from learned patterns and supplied information.
3. Numeric regression output

   Regression predicts numeric values and does not directly produce fluent product descriptions.
4. Predictive category output

   Category prediction assigns labels, such as departments, rather than composing new product descriptions.

Generating product descriptions is a generative task because the system creates new language from learned patterns.

4. An AI agent: Which capability best fits?

1. An AI agent ✓

   An agent can use tools and perform multiple steps toward a goal, rather than only producing one response.
2. A single text-generation response

   A single response can generate prose but does not inherently coordinate searches, comparisons, and subsequent actions.
3. A clustering model

   Clustering groups similar observations and does not inherently search sources or draft reports.
4. A classification model

   Classification assigns categories and does not inherently conduct multi-step research activities.

Agents can use tools and coordinate multiple actions toward goals, matching the research workflow described.

5. Running inference with the trained model: Which activity is required?

1. Retraining the model with new parameters

   Retraining changes learned parameters and is unnecessary when applying an already trained model.
2. Preparing labeled examples for supervised learning

   Labeled examples support training, but the requirement concerns scoring current transactions.
3. Running inference with the trained model ✓

   Inference applies a trained model to new inputs and produces predictions for immediate recommendations.
4. Discovering unlabeled groups among historical transactions

   Unsupervised grouping analyzes structure and does not directly score current transactions with the trained model.

Inference applies an already trained model to new transaction data and produces the requested recommendations.

6. Supervised learning that learns from examples containing: Which learning approach best uses these examples?

1. Supervised learning that learns from examples containing known target answers ✓

   Supervised learning uses labeled examples with known targets, matching translations and expert quality scores.
2. Unsupervised learning that discovers structure without target labels

   Unsupervised learning does not use supplied target labels, unlike the paired translations and expert scores.
3. Clustering that groups exercises according to similarity patterns

   Clustering groups observations without supplied correct answers and does not directly learn the labeled translation target.
4. Reinforcement learning that learns only from delayed environmental rewards

   Reinforcement learning learns behavior through rewards, while this dataset directly supplies desired outputs.

Supervised learning is appropriate because each exercise includes known target translations or expert quality labels.

7. Clustering: Which approach is best?

1. Classification using predefined customer categories

   Classification requires known categories or labeled examples, which the marketplace explicitly lacks.
2. Regression predicting customer spending amounts

   Regression predicts numeric values and does not primarily discover naturally similar customer groups.
3. Reinforcement learning using purchase rewards

   Reinforcement learning learns behavior through rewards, not simply customer similarity from existing records.
4. Clustering ✓

   Clustering groups similar observations without requiring supplied class labels or predefined segments.

Clustering discovers similar groups in data without requiring predefined labels or customer segments.

8. Reinforcement learning: Which approach fits?

1. Supervised learning from fixed answer labels

   Supervised learning uses labeled targets, whereas this requirement emphasizes action sequences and outcome feedback.
2. Reinforcement learning ✓

   Reinforcement learning learns behavior through rewards or feedback associated with actions and resulting outcomes.
3. Clustering documents by visual similarity

   Clustering discovers groups of similar documents but does not learn action choices from processing outcomes.
4. Regression predicting processing duration only

   Regression can estimate numeric duration but does not inherently learn which actions improve outcomes.

Reinforcement learning learns action choices through feedback or rewards tied to resulting processing outcomes.

9. Preserving the sequence of observations over time: What characteristic is essential?

1. Assigning each observation to an unrelated category

   Categorization may organize records but does not preserve the temporal order required for forecasting.
2. Grouping observations without considering their order

   Ignoring order removes the temporal relationships needed to forecast future volume accurately.
3. Preserving the sequence of observations over time ✓

   Time series data depends on ordered observations because timing and prior values influence later forecasts.
4. Replacing historical observations with generated text

   Generated text does not preserve the numeric temporal relationships required for volume forecasting.

Time series forecasting requires preserving observation order because timing and prior values influence later predictions.

10. The cancellation outcome is the label: Which statement correctly identifies the label and features?

1. The cancellation outcome is the label, while timing and destination are features. ✓

   The label is the target to predict, and reservation attributes are features used to estimate it.
2. Reservation details form unlabeled groups, and cancellations are clusters.

   The scenario describes supervised prediction, not unsupervised grouping into clusters.
3. The cancellation outcome is a feature, and reservation details are labels.

   The outcome being predicted is the label; reservation details provide input features.
4. Both the cancellation outcome and reservation details are labels.

   The cancellation outcome is the target label, while reservation details are typically input features.

The label is the target outcome; features are input attributes used to predict it.

11. Use batch inference to score the monthly donor collection: Which approach fits best?

1. Use asynchronous inference to queue each donor record for later scoring.

   Asynchronous inference can delay individual requests, but the requirement is a scheduled collection-wide workload.
2. Use clustering to organize donors into groups without individual scores.

   Clustering finds groups and does not produce the requested score for every donor record.
3. Use batch inference to score the monthly donor collection overnight. ✓

   Batch inference processes a collection when immediate responses are not required, matching the monthly schedule.
4. Use real-time inference as each donor record becomes available.

   Real-time inference is intended for interactive, low-latency requests, which are unnecessary here.

Batch inference fits scheduled collection processing when immediate responses are unnecessary.

12. Real-time inference for interactive requests requiring: Which inference approach is most appropriate?

1. Real-time inference for interactive requests requiring prompt responses. ✓

   Real-time inference is intended for interactive requests where users expect responses during the conversation.
2. Asynchronous inference for queued questions with delayed responses.

   Asynchronous inference accepts delayed results, conflicting with the help desk's interactive response expectation.
3. Batch inference for scheduled collections of employee questions.

   Batch inference is designed for collections processed together, not conversations requiring answers during interaction.
4. Clustering inference to organize employee questions by similarity.

   Clustering can organize similar items, but it does not itself provide conversational responses to employees.

Real-time inference best supports interactive conversations where users need prompt responses during each exchange.

Other AWS AI Practitioner domains

• Applications of Foundation Models — 311 questions →
• Fundamentals of GenAI — 269 questions →
• Security, Compliance, and Governance for AI Solutions — 157 questions →
• Guidelines for Responsible AI — 156 questions →
• All 1116 AWS AI Practitioner questions →