Importance of transparent and explainable mod..

Option A is correct because Amazon SageMaker Model Cards provide a structured location for documenting model intended use, training details, evaluation results, risk information, limitations, and other governance considerations. Option B is correct because documenting data sources, preprocessing, provenance, and licensing terms supports responsible data governance and helps establish how the training data was obtained and used. Option C is a computer vision model-training capability rather than a model-governance tool. Option D is an operational data-storage choice, and Option E addresses training performance rather than transparency or governance.

Option A is correct because human-centered explainability should present relevant model information in a form that users can understand and use when making decisions. A clear dashboard with key contributing factors can help managers review and appropriately act on recommendations. Option B provides technical logs that may be difficult for operational users to interpret. Options C and D reduce user visibility, feedback, and appropriate human oversight.

Option D is correct because the firm must balance strong detection performance with the need for understandable alerts that analysts can investigate and communicate effectively. A low false-negative rate indicates strong sensitivity to threats but does not explain why a particular alert was generated. Options A, B, and C address data, infrastructure, or licensing considerations rather than the interpretability trade-off.

Option D is correct because a more complex model may provide stronger predictive performance for a task with complex patterns, while an inherently interpretable linear model provides clearer coefficients and a simpler decision structure. The team should evaluate whether any performance improvement justifies the additional complexity and reduced interpretability. Option A is incorrect because linear models can operate on transformed representations such as embeddings. Option B incorrectly claims that complex neural networks require fewer resources, and Option C incorrectly claims that linear models cannot run on serverless infrastructure.

Option A is correct because model explainability helps operations managers understand which input factors contribute to predictions and can support informed business decisions. Hardware acceleration (Option B), training frequency (Option C), and storage optimization (Option D) can affect system performance or operations but do not directly explain model predictions.

Option B is correct because model transparency supports the ability to communicate the underlying decision logic in a way that is understandable to users. Maximum parameter count (Option A) and high memory consumption (Option D) are not transparency characteristics. Proprietary algorithm secrecy (Option C) may protect intellectual property, but it does not itself provide the user-facing transparency required by the scenario.

Option B is correct because post-hoc explanation techniques can provide understandable explanations for recommendations produced by a high-performance model. Option C is correct because an approach that uses interpretable item or user attributes can provide more direct explanations while still requiring evaluation of recommendation quality. Option A may provide some transparency but does not necessarily provide the predictive performance or recommendation capability required for a large product catalog. Option D does not address the transparency requirement. Option E prioritizes model simplicity without establishing that the model can provide adequate recommendation quality.

Option A is correct because Amazon Bedrock Model Evaluations supports automatic evaluations and human-based evaluations for supported models and resources, helping teams assess model effectiveness using defined metrics and evaluation criteria. Option B is correct because accessible model implementations combined with appropriate explainability or feature-attribution techniques can help teams investigate how recommendations are produced. Option C is primarily a data-labeling capability, while Option D focuses on data streaming and Option E focuses on deployment performance and latency rather than model transparency or explainability.