PMI-CPMAI Sample Questions & Answers

The F-beta score is a generalization of the F1-score that allows for weighting precision over recall, or vice versa. A beta value less than 1 (e.g., F0.5-score) gives more weight to precision (minimizing False Positives), while a beta greater than 1 gives more weight to recall (minimizing False Negatives). Since a default (False Positive) is five times more costly, the model should prioritize precision. Therefore, an F-beta score with beta < 1 is the most appropriate metric. The F1-score gives equal weight to both. AUC is a general measure of separability and does not directly account for asymmetric costs.

Data lineage is the process of understanding, recording, and visualizing data as it flows from data sources to consumption. This includes all transformations the data underwent along the way. Implementing robust data lineage is essential for meeting the requirements of traceability, auditing, and debugging AI models. Data masking and anonymization are privacy techniques, and data augmentation is a method for increasing training data size.

A feature store is a central repository for features used in machine learning. Its primary purpose is to solve the problem of feature consistency between training and serving, reduce duplicate feature engineering work across teams, and provide a reliable, low-latency source of feature data for real-time inference. This centralization is a key component of a mature MLOps strategy.

With limited specific data, training an LLM from scratch is infeasible. The most effective strategy is to leverage a powerful pre-trained model and augment its knowledge with domain-specific information. Retrieval-Augmented Generation (RAG) is designed for this; it retrieves relevant information from a knowledge base (like product documentation) and provides it to the LLM as context to generate an accurate answer. This is faster, cheaper, and more effective than fine-tuning with insufficient data or building from scratch.

This is the most accurate description. SHAP (SHapley Additive exPlanations) is a game theory-based approach to explain the output of any machine learning model. It connects optimal credit allocation with local explanations using the classic Shapley values. For a specific prediction, it shows which features pushed the prediction higher or lower and by how much, providing crucial local interpretability for 'black box' models.

Concept drift occurs when the statistical properties of the target variable change over time. For churn, this could mean the underlying reasons why customers leave have changed (e.g., a new competitor entered the market, changing customer behavior).

Data drift (or feature drift) happens when the statistical properties of the input features change. For example, the average customer tenure or product usage patterns might have shifted since the model was trained, making the production data different from the training data.

AI projects are inherently multi-disciplinary. A cross-functional team structure that brings together all necessary skills (data engineering, machine learning science, operations/deployment, and business context) fosters collaboration, reduces hand-off friction, and ensures the end product is aligned with business needs and technically robust. Siloed approaches (A) are inefficient and prone to error. A team of only data scientists (B) would lack critical engineering and business skills. Outsourcing core development (D) can work but is not inherently the most successful structure.

Data augmentation is a set of techniques used to increase the amount of data by adding slightly modified copies of already existing data or newly created synthetic data. For images, this involves operations like rotation, cropping, flipping, and adjusting brightness/contrast. This allows the model to learn from a wider variety of examples without the high cost of collecting and labeling new, unique images, making it the most cost-effective solution.

The correct hierarchy is that Artificial Intelligence (AI) is the broad field, Machine Learning (ML) is a subset of AI, and Deep Learning (DL) is a further subset of ML. Deep Learning utilizes multi-layered artificial neural networks (hence 'deep') to learn from vast amounts of data, enabling it to solve more complex problems than some traditional ML algorithms.

Accuracy is the most misleading metric in cases of severe class imbalance. A naive model that simply classifies every ticket as 'Billing' would achieve 80% accuracy but would be completely useless for identifying 'Technical Issue' or 'Feedback' tickets. Metrics like Precision, Recall, and especially the F1-score (or a macro-averaged score) provide a much better assessment of performance across all classes, including the rare ones.