View Full Amazon AWS Certified AI Practitioner AIF-C01 Exam Dumps and Practice Test Dumps.

 

Question 221

Which machine learning technique is used to identify natural groupings within a dataset when the groups are not predefined?

1. Clustering
2. Classification
3. Regression
4. Reinforcement learning

Correct Answer: 1

Explanation

Clustering is an unsupervised learning technique that groups data points according to similarities or patterns without requiring predefined labels. It can be useful for customer segmentation, document grouping, anomaly investigation, or exploratory data analysis. The algorithm attempts to identify meaningful structures within the available data rather than learning from known target categories. Different clustering methods use different definitions of similarity and may produce different groupings. Organizations should evaluate whether the resulting clusters have meaningful business interpretations before using them for decisions or downstream processes.

Question 222

A data scientist wants to determine how much each input variable contributes to a model’s prediction. Which concept is most directly related to this goal?

1. Data residency
2. Feature importance
3. Tokenization
4. Model quantization

Correct Answer: 2

Explanation

Feature importance describes the relative contribution or influence of input features on a model’s predictions, depending on the model and interpretation technique being used. Understanding feature importance can help teams investigate model behavior, identify influential variables, and detect potentially problematic relationships. However, feature importance should not automatically be interpreted as proof of causation. Different models and explanation methods can produce different interpretations. Organizations should consider the model type, evaluation method, and application context when interpreting feature importance, particularly when predictions are used in sensitive or high-impact workflows.

Question 223

Which situation is an example of regression rather than classification?

1. Determining whether a transaction is fraudulent
2. Predicting whether a customer will cancel a service
3. Estimating the future monthly electricity consumption of a building
4. Determining whether an image contains a vehicle

Correct Answer: 3

Explanation

Regression predicts a continuous numerical value rather than selecting a discrete category. Estimating future electricity consumption is therefore a regression task because the output could be a numerical quantity such as kilowatt-hours. Classification, by contrast, predicts categories such as fraudulent or legitimate, canceled or retained, or vehicle or no vehicle. The choice between regression and classification depends on the type of target variable and the intended prediction. Regression models should be evaluated using suitable numerical metrics and representative data that reflects the conditions under which predictions will be made.

Question 224

A company wants to determine whether an AI system consistently identifies objects in images despite changes in lighting and image quality. What type of evaluation is most relevant?

1. Cost analysis
2. Storage testing
3. Permission testing
4. Robustness testing

Correct Answer: 4

Explanation

Robustness testing examines whether an AI system maintains appropriate performance when inputs vary from ideal conditions. For an image recognition system, changes in lighting, resolution, background, camera angle, or image quality may reveal weaknesses that are not visible in standard testing. Evaluating these variations can help determine whether the system is reliable under realistic operating conditions. The specific tests should reflect the application’s expected environment. Robustness testing should complement normal accuracy evaluation because an overall performance score may not reveal failures associated with particular input conditions.

Question 225

Which practice can help reduce the risk that sensitive information is unintentionally included in prompts sent to a generative AI model?

1. Applying data filtering or redaction before sending the prompt
2. Increasing the model temperature
3. Expanding the output token limit
4. Disabling authentication

Correct Answer: 1

Explanation

Filtering or redacting sensitive information before sending prompts can reduce the risk of unintentionally exposing confidential data to a model. Organizations should first determine what information is necessary for the requested task and remove unnecessary sensitive fields where possible. Depending on the use case, techniques such as masking, pseudonymization, or automated PII detection may be appropriate. Prompt protection should be combined with access controls, secure transmission, logging policies, and appropriate data-retention practices. Redaction strategies should also be tested to ensure that removing sensitive information does not make the application unusable.

Question 226

A model’s training dataset contains many more examples from one class than another. What issue should the team investigate?

1. Model quantization
2. Class imbalance
3. Context overflow
4. Token sampling

Correct Answer: 2

Explanation

Class imbalance occurs when one or more classes have substantially more examples than other classes in a dataset. This can cause a classification model to favor the majority class and perform poorly on less represented categories. Accuracy alone may hide this problem, particularly when the majority class dominates the dataset. Teams can investigate metrics such as precision, recall, and F1 score and consider appropriate data or modeling strategies. The correct response depends on the application and data. Evaluation should specifically examine performance across relevant classes rather than relying only on aggregate results.

Question 227

An organization needs to convert scanned forms into machine-readable text and extract fields such as names and dates. Which AWS service is designed for this type of workload?

1. Amazon Transcribe
2. Amazon Translate
3. Amazon Textract
4. Amazon Polly

Correct Answer: 3

Explanation

Amazon Textract is designed to extract text and structured information from documents. It can process scanned forms and other documents and identify information such as text, forms, and tables. This makes it useful for workflows involving invoices, applications, financial documents, and other structured or semi-structured content. Textract differs from Amazon Transcribe, which focuses on speech-to-text, and Amazon Translate, which translates languages. Organizations should evaluate extraction quality using representative documents because scan quality, layout complexity, handwriting, and document variations can affect results.

Question 228

Which approach can help an AI application avoid repeatedly sending the same large prompt or context when supported by the selected service?

1. Increasing model size
2. Prompt caching
3. Changing the database schema
4. Disabling monitoring

Correct Answer: 2

Explanation

Prompt caching can help reduce repeated processing of identical or reusable prompt content when the selected model and service support this capability. This can be useful for applications that repeatedly send a large static system instruction, document context, or other reusable information. Potential benefits can include reduced latency or cost depending on the service’s pricing and implementation. Caching must be designed carefully because cached information may become outdated or inappropriate for a particular request. Organizations should verify service-specific caching behavior, eligibility, retention, and pricing before incorporating it into an application architecture.

Question 229

An AI application must select an appropriate tool based on a user’s request and then use the tool’s result to continue the task. Which agent capability is most relevant?

1. Tool selection and orchestration
2. Data encryption
3. Model compression
4. Feature normalization

Correct Answer: 1

Explanation

Tool selection and orchestration allow an AI agent to determine which available capability can help complete a user’s request and coordinate the resulting workflow. For example, an agent may decide whether it needs to retrieve information, query a business system, or perform another supported operation. The model should not receive unrestricted authority simply because it can select tools. Applications should enforce authentication, authorization, input validation, and action-specific safeguards. Proper orchestration separates the agent’s reasoning and tool-selection behavior from the controlled execution of external operations.

Question 230

Which characteristic of a dataset is especially important when training a model intended for use across different demographic groups?

1. Low storage capacity
2. Representative diversity
3. Random file naming
4. Maximum compression

Correct Answer: 2

Explanation

Representative diversity is important when a model is intended to serve users from different demographic groups or operating conditions. A dataset that adequately represents relevant populations can help the model learn patterns that are applicable across those groups. Insufficient representation can contribute to uneven performance or unwanted bias. Dataset diversity should be considered alongside data quality, labeling accuracy, relevance, and the actual population expected during deployment. Organizations should also evaluate model performance across relevant groups rather than assuming that a diverse training dataset automatically guarantees fair outcomes.

Question 231

A company wants to determine whether its model’s predictions change substantially when small, reasonable variations are made to otherwise similar inputs. What should it evaluate?

1. Model sensitivity and robustness
2. Storage encryption
3. User authentication
4. Data residency

Correct Answer: 1

Explanation

Evaluating model sensitivity and robustness can reveal whether small and reasonable input changes cause inappropriate or unexpected changes in predictions. This type of testing can be useful for identifying models that are overly sensitive to irrelevant variations. For example, an application may test slight changes in wording, image conditions, or numerical inputs while keeping the underlying meaning unchanged. The appropriate tolerance depends on the use case. Results should be interpreted alongside accuracy, safety, fairness, and other evaluation dimensions to determine whether the observed behavior is acceptable.

Question 232

Which AWS service is primarily designed to provide scalable object storage that can be used for datasets, documents, and model artifacts?

1. Amazon SQS
2. Amazon S3
3. Amazon VPC
4. Amazon Route 53

Correct Answer: 2

Explanation

Amazon S3 provides scalable object storage that can be used to store datasets, documents, model artifacts, logs, and other files used by AI and machine learning workloads. S3 supports features for access management, lifecycle policies, encryption, versioning, and other storage requirements. AI applications frequently use S3 as a durable storage layer for training and evaluation data or supporting documents. Organizations should configure appropriate permissions and security controls because storing data in S3 does not automatically make it accessible only to authorized users.

Question 233

A team wants to identify whether an AI model produces systematically different error rates for relevant groups. Which analysis is most appropriate?

1. Fairness analysis
2. DNS analysis
3. Storage analysis
4. Network routing analysis

Correct Answer: 1

Explanation

Fairness analysis can help identify whether a model behaves differently across relevant groups, including whether error rates or other outcomes differ systematically. The appropriate fairness criteria depend on the application’s purpose, population, and potential impact. Teams should define which groups and outcomes are relevant before performing the analysis. Differences do not automatically establish a cause, so organizations may need additional investigation into data quality, sampling, features, or model behavior. Fairness assessment is part of responsible AI evaluation and should be considered alongside accuracy, robustness, privacy, safety, and other requirements.

Question 234

A model generates technically valid text but repeatedly ignores an application’s required response structure. Which improvement is most directly relevant?

1. Increase network bandwidth
2. Strengthen prompt instructions and output constraints
3. Change the storage class
4. Disable model evaluation

Correct Answer: 2

Explanation

Clear prompt instructions and output constraints can help a generative model follow a required response structure. The prompt can explicitly define the expected fields, ordering, formatting, and restrictions. Providing examples may also help communicate the desired structure when appropriate. However, generated output should still be validated by the application because a model can occasionally produce malformed or incomplete responses. If strict machine-readable behavior is required, application-level schema validation should complement prompting. Changing unrelated infrastructure settings would not directly address a model’s tendency to ignore the requested output format.

Question 235

Which metric measures the proportion of predicted positive results that are actually positive?

1. Recall
2. Accuracy
3. Precision
4. Mean absolute error

Correct Answer: 3

Explanation

Precision measures the proportion of predicted positive results that are actually positive. It is calculated as true positives divided by the sum of true positives and false positives. Precision is useful when false positive predictions are particularly important to the application. For example, a system that flags transactions as suspicious may need to limit the number of legitimate transactions incorrectly flagged. Precision differs from recall, which measures how many actual positive cases were successfully identified. Organizations should select evaluation metrics according to the consequences of different types of errors in the specific application.

Question 236

A model evaluation requires comparing predicted numerical values with actual numerical values. Which metric measures the average magnitude of prediction errors without squaring them?

1. Accuracy
2. Mean absolute error
3. Recall
4. Precision

Correct Answer: 2

Explanation

Mean absolute error, or MAE, measures the average absolute difference between predicted numerical values and their corresponding actual values. Because the errors are converted to absolute values before averaging, positive and negative errors do not cancel each other out. MAE is commonly used for regression problems and is expressed in the same units as the target variable. This can make it relatively intuitive to interpret. Other metrics, such as mean squared error, penalize larger errors more strongly. The appropriate metric depends on the business consequences and evaluation requirements of the regression task.

Question 237

Which approach is most useful when an AI application needs to search documents based on meaning rather than requiring an exact keyword match?

1. Semantic search
2. File compression
3. DNS lookup
4. Data encryption

Correct Answer: 1

Explanation

Semantic search retrieves information based on the meaning or conceptual similarity of a query rather than requiring exact keyword matches. It commonly uses vector representations or embeddings to compare the meaning of queries and documents. This can help retrieve relevant information when the wording used by the user differs from the wording in the source material. Semantic search can be combined with keyword search in a hybrid approach. Retrieval quality should be evaluated with representative queries because semantic similarity alone does not guarantee that the retrieved document is the most useful source for a particular question.

Question 238

A company wants to determine whether a model’s performance remains acceptable after a new version of the model is introduced. What should the team perform?

1. Delete the previous model immediately
2. Compare the new version against defined evaluation criteria
3. Disable monitoring
4. Grant the new model unrestricted permissions

Correct Answer: 2

Explanation

A new model version should be evaluated against defined criteria before or during controlled deployment. The team can compare relevant quality, safety, performance, cost, latency, and robustness measurements with established requirements or an existing baseline. Using consistent evaluation procedures helps make the comparison meaningful. The previous model should not necessarily be deleted immediately because it may be needed for rollback or comparison. Organizations should also maintain version information and deployment records. Model updates can change behavior in unexpected ways, so testing should use representative production-like inputs.

Question 239

Which AWS service can convert synthesized text into natural-sounding speech for applications such as voice assistants?

1. Amazon Polly
2. Amazon Macie
3. Amazon Kendra
4. Amazon SageMaker

Correct Answer: 1

Explanation

Amazon Polly is an AWS service that converts text into spoken audio using speech synthesis capabilities. It can be used in applications such as voice assistants, accessibility solutions, announcements, and other systems that need generated speech. The service supports different voices and languages, depending on the capabilities available for the selected configuration. Organizations should evaluate pronunciation, language support, voice characteristics, latency, and output quality for their particular use case. Polly is distinct from Amazon Transcribe, which converts spoken audio into text rather than generating speech from text.

Question 240

A team wants to ensure that an AI application can explain which source documents contributed to a generated answer. Which design practice can improve this capability?

1. Remove document metadata
2. Preserve source references and provide citations or attribution with retrieved information
3. Disable retrieval
4. Hide all source information from the application

Correct Answer: 2

Explanation

Preserving source references and providing citations or attribution can improve traceability in applications that generate answers from retrieved information. When a response is grounded in documents, maintaining metadata about the source can help users verify where relevant information came from. This can also support troubleshooting and governance by allowing teams to investigate the material used to construct an answer. Source attribution does not guarantee that a generated response is correct, so organizations should still evaluate retrieval quality and model behavior. Appropriate access controls must also be maintained when exposing source information to users.