AIF-C01 Practice Exam — AIF-C01：AI Practitioner

Option Analysis: A. Option A is incorrect because providing full model training code is unnecessary for non-technical stakeholder transparency and does not help stakeholders understand how the model makes predictions. B. Option B is correct because partial dependence plots (PDPs) are explainable AI tools that clearly show how input features impact forecast output, making them ideal for communicating model behavior to stakeholders to meet transparency requirements. C. Option C is incorrect because including sample training data does not explain how the trained model behaves or makes predictions, so it does not meet transparency or explainability requirements. D. Option D is incorrect because model convergence tables only show training performance metrics over epochs, and do not explain how the model's input features affect its forecasts for stakeholders. Key Concept: Explainable AI (XAI) techniques for machine learning help provide transparency into model prediction behavior for non-technical stakeholders. References: What Is Explainable AI (XAI)?, Amazon SageMaker Clarify, https://docs.aws.amazon.com/sagemaker/latest/dg/clarify.html

Option Analysis: A. Option A is incorrect because automatic named entity recognition only identifies specific predefined entities like people or organizations, it does not extract broad key points from entire legal documents as required. B. Option B is incorrect because a recommendation engine is designed to suggest relevant content to users, it is not built to extract key points from input documents. C. Option C is correct because LLM-based summarization applications are purpose-built to condense full length documents into their core key points, which matches the law firm's requirement. D. Option D is incorrect because a multi-language translation system only converts text between different languages, it does not perform key point extraction from source documents. Key Concept: Large language model (LLM) summarization use cases focus on condensing long-form text into concise, relevant key takeaways for document processing workflows. References: What Is Text Summarization?, AWS docs, Foundation Models Use Cases, AWS docs, https://docs.aws.amazon.com/sagemaker/latest/dg/fm-use-cases.html

Option Analysis: A. Decision trees are correct because they naturally handle multi-class classification (the 20 gene categories) and produce fully interpretable, step-by-step decision paths that clearly document how input characteristics lead to the final output, meeting the explainability requirement. B. Linear regression is incorrect because it is an algorithm for regression (predicting continuous values) not classification of categorical outputs. C. Logistic regression is incorrect because while it can be adapted for multi-class classification, it does not provide the clear, granular inner mechanism documentation that decision trees offer for how features impact predictions. D. Neural networks are incorrect because they are generally black box models with opaque inner mechanisms that are difficult to interpret and document for how inputs affect outputs. Key Concept: Interpretability vs. opacity of machine learning algorithms, where tree-based models provide full transparency into a model's decision making process for classification tasks. References: What Is Machine Learning? - AWS Machine Learning, https://docs.aws.amazon.com/machine-learning/latest/dg/what-is-amazon-machine-learning.html Decision Trees - AWS Machine Learning Algorithm Reference, https://docs.aws.amazon.com/machine-learning/latest/dg/decision-trees.html

Option Analysis: A. Option A is incorrect because R-squared score is an evaluation metric for regression models that measures how well the independent variable explains the variance of the dependent variable, not the count of correct classifications for classification tasks. B. Option B is correct because accuracy directly measures the proportion (and corresponding count) of correctly classified predictions out of all total predictions, which matches the requirement to count how many images the model classified correctly. C. Option C is incorrect because Root mean squared error is a regression evaluation metric that measures the average prediction error magnitude, it is not used to count correct classification results. D. Option D is incorrect because learning rate is a hyperparameter that controls how much model weights are updated during training, it is not an evaluation metric for model performance. Key Concept: Accuracy is the primary classification evaluation metric that calculates the share of correct predictions out of all predictions, used to measure overall classification performance when the question asks for the count of correct classifications. References: Classification Metrics for Machine Learning, https://docs.aws.amazon.com/machine-learning/latest/dg/evaluation-metrics-for-classification.html Amazon SageMaker Model Evaluation Metrics, https://docs.aws.amazon.com/sagemaker/latest/dg/model-evaluation-metrics.html

Option Analysis: A. This option is correct because prompt engineering, specifically explicitly stating requirements for short output length and specific language directly in the prompt, is the most straightforward and effective way to align LLM output with these expectations. B. This option is incorrect because changing the LLM size does not reliably enforce the specific output formatting and language requirements the company needs, it only changes the model's general capability and capacity. C. This option is incorrect because increasing temperature makes LLM output more random and creative, which does not help enforce short output or specific language requirements. D. This option is incorrect because increasing Top K value allows the model to select from a larger set of possible next tokens, which increases output randomness and does not enforce the required output constraints. Key Concept: Prompt engineering is the practice of designing input prompts to guide a large language model to produce desired output that aligns with specific requirements. References: Prompt engineering for foundation models, https://docs.aws.amazon.com/sagemaker/latest/dg/jumpstart-foundation-models-prompt-engineering.html Working with LLMs on Amazon Bedrock, https://docs.aws.amazon.com/bedrock/latest/userguide/prompt-engineering.html